{
  "schema_version": "1.0",
  "generated_at": "2026-10-30T15:09:49.951Z",
  "evidence_note": "Vendor documentation researched, not hands-on tested. Unknown values are not free or absent features. Record last_checked is not the API generation time.",
  "documentation_url": "https://agentcomparison.net/api/",
  "count": 77,
  "agents": [
    {
      "id": "chatgpt-workspace-agents",
      "name": "ChatGPT workspace agents",
      "vendor": "OpenAI",
      "kind": "platform",
      "category": "workspace workflow agents",
      "group": "work-business",
      "summary": "OpenAI documents repeatable workflows in ChatGPT and Slack, with connected tools, schedules, API triggers and workspace access controls.",
      "one_liner": "Repeatable team workflows in ChatGPT and Slack, with admin controls.",
      "best_fit": "Repeatable team workflows with configured access and triggers.",
      "poor_fit": "Assuming every ChatGPT subscription includes workspace agents.",
      "decision_note": "Verify plan eligibility and admin enablement before recommending; do not conflate this with historical ChatGPT agent mode.",
      "availability": "Main help guide covers Business and Enterprise workspaces; administrator enablement and role permissions apply. Do not assume every ChatGPT plan includes workspace agents.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Schedules and API triggers",
          "source_url": "https://help.openai.com/en/articles/20001143-chatgpt-workspace-agents-for-enterprise-and-business",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Apps and Slack channels",
          "source_url": "https://help.openai.com/en/articles/20001143-chatgpt-workspace-agents-for-enterprise-and-business",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Write-action approvals and parameter constraints",
          "source_url": "https://help.openai.com/en/articles/20001143-chatgpt-workspace-agents-for-enterprise-and-business",
          "observed_at": "2026-10-10",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Multiplayer editing and group sharing",
          "source_url": "https://help.openai.com/en/articles/20001143-chatgpt-workspace-agents-for-enterprise-and-business",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Business and Enterprise workspaces",
          "region_restrictions": null
        }
      ],
      "permissions": [
        "End-user authentication runs through each person's account; agent-owned authentication shares a connection. Use a service account with minimum access.",
        "Write actions default to Always ask; Never ask or Custom settings may be available by app. Review risky send/edit/delete workflows.",
        "Connector action constraints narrow action inputs, not the data returned.",
        "Review private, invited and workspace sharing audiences before publication. Slack requires shared app authentication."
      ],
      "limitations": [
        "API trigger queues a run with 202 Accepted; the JSON body carries conversation_url (which can be null), and a beta-header flow adds agent_trigger_run_id for status polling. Retrieving the final response through that API remains limited.",
        "Connector constraints limit requested actions, not returned data.",
        "Slack requires shared authentication for every app connection; personal shared accounts can expose owner data/actions to other users.",
        "A separate workspace-agents security overview was referenced, but not established in this review. No product-specific retention or independent certification claim is made."
      ],
      "runtime": [
        "Supported ChatGPT workspaces with agents enabled"
      ],
      "privacy_security": null,
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-22",
        "source_url": "https://help.openai.com/en/articles/11481834",
        "note": "OpenAI's credit rate card for Business/Enterprise flexible pricing bills Workspace Agents by tokens: GPT-5.6 at 100 credits per 1M input, 10 per 1M cached input, 500 per 1M output; OpenAI says a typical end-to-end run uses 5-25 credits. Workspace Agents draw from the same agentic usage and credit pool as Codex, ChatGPT Work, Excel and PowerPoint; standard Business seats include usage from that pool, with flexible pricing extending limits."
      },
      "sources": [
        "https://help.openai.com/en/articles/20001143-chatgpt-workspace-agents-for-enterprise-and-business",
        "https://help.openai.com/en/articles/11481834",
        "https://help.openai.com/en/articles/20001155"
      ],
      "changelog": [
        {
          "date": "2026-10-10",
          "change": "Expanded source-linked feature evidence and narrow setup; clarified access, permissions, costs and unresolved limits. Documentation research only, not a product test."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        }
      ],
      "coverage_rows": [
        [
          "Reusable workflows",
          "Documented",
          "Create from a template or builder, choose tools/model, preview with a sample task, then publish. Workspace eligibility and administrator controls apply.",
          "https://help.openai.com/en/articles/20001143-chatgpt-workspace-agents-for-enterprise-and-business"
        ],
        [
          "Tools and apps",
          "Documented",
          "Google Calendar, Drive, Slack, SharePoint, custom MCPs, web search and image generation are documented. Availability depends on workspace enablement.",
          "https://help.openai.com/en/articles/20001143-chatgpt-workspace-agents-for-enterprise-and-business"
        ],
        [
          "End-user versus shared authentication",
          "Documented",
          "End-user connections run as each person; agent-owned connections are shared. A personal shared connection can expose its owner's data and actions to others. OpenAI recommends service accounts and least access.",
          "https://help.openai.com/en/articles/20001143-chatgpt-workspace-agents-for-enterprise-and-business"
        ],
        [
          "Write approvals",
          "Configurable",
          "Apps/connectors default to Always ask. Depending on the app, Never ask or Custom controls may be available. Do not claim every write always asks.",
          "https://help.openai.com/en/articles/20001143-chatgpt-workspace-agents-for-enterprise-and-business"
        ],
        [
          "Action constraints",
          "Not a data filter",
          "Constraints limit what the agent asks a connector to do. They do not filter what the connector returns, so an allowed search may still return sensitive data.",
          "https://help.openai.com/en/articles/20001143-chatgpt-workspace-agents-for-enterprise-and-business"
        ],
        [
          "Schedules and API triggers",
          "Documented with limits",
          "Schedules and API channels are supported. API queues a run with 202 Accepted, no response body or run ID; the agent response cannot currently be retrieved through that API.",
          "https://help.openai.com/en/articles/20001143-chatgpt-workspace-agents-for-enterprise-and-business"
        ],
        [
          "Slack channels",
          "Shared connections required",
          "All app connections must use shared authentication for Slack. Choose every-message or mention-only responses; review the channel audience and credentials before deployment.",
          "https://help.openai.com/en/articles/20001143-chatgpt-workspace-agents-for-enterprise-and-business"
        ],
        [
          "Business usage and credits",
          "Plan-dependent",
          "Current Business billing docs describe included seat usage followed by workspace credits for eligible additional usage and spend controls. No universal per-agent run price is claimed.",
          "https://help.openai.com/en/articles/20001155"
        ]
      ],
      "setup_steps": [
        "Confirm Agents is available and enabled in the intended workspace.",
        "Build one narrow workflow and keep it private while previewing a sample task.",
        "Choose end-user or service-account shared authentication deliberately; keep write approval on for risky actions.",
        "Review outputs and access before publishing, adding schedules or Slack. Confirm eligible usage and spend controls with the workspace owner."
      ],
      "price_tags": [
        "usage"
      ],
      "price_summary": "Token-based workspace credits on Business/Enterprise flexible pricing; OpenAI rate card: typical GPT-5.6 agent run 5-25 credits",
      "meta_title": "ChatGPT workspace agents: plans, setup and controls",
      "page_url": "https://agentcomparison.net/agents/chatgpt-workspace-agents/"
    },
    {
      "id": "claude-code",
      "name": "Claude Code",
      "vendor": "Anthropic",
      "kind": "product",
      "category": "coding agent",
      "group": "coding-agents",
      "summary": "Claude Code reads repositories, edits files, runs commands and connects development tools through terminal, IDE, desktop and web surfaces. Its current security docs describe Auto mode as the starting mode for interactive terminal and VS Code sessions, while Manual, Plan and other modes change approvals. Local execution does not mean code and prompts stay offline.",
      "one_liner": "Anthropic's terminal coding agent for migrations, bug fixes and refactors.",
      "best_fit": "Developers delegating repository changes from a terminal.",
      "poor_fit": "A no-setup personal assistant for errands.",
      "decision_note": "Start from the actual permission mode, repository rules and account data settings, not the older claim that every edit asks first. Use a branch and your own tests. Optional sandboxing gives different boundaries from a prompt-only working directory; installed extensions and cloud sessions need separate review.",
      "availability": "Terminal, VS Code, JetBrains, desktop and web are documented. Most surfaces require a Claude subscription or Console account; terminal and IDEs also support third-party providers. Pricing table excludes Code from Free. Regional and team seat eligibility should be confirmed before use.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Repository work: Reads a codebase, edits multiple files, runs tests/commands and documents bugs or features. Generated code still needs review and your test suite.",
          "source_url": "https://code.claude.com/docs/en/overview",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Account, session mode and policy dependent",
          "region_restrictions": "Not established"
        },
        {
          "name": "Git and CI: Branches, commits and pull requests; GitHub Actions and GitLab CI/CD for reviews/triage. A repository change is not automatically a safe merge.",
          "source_url": "https://code.claude.com/docs/en/overview",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Account, session mode and policy dependent",
          "region_restrictions": "Not established"
        },
        {
          "name": "Where it runs: Terminal, VS Code/JetBrains, desktop and web. Local CLI and Anthropic-hosted cloud sessions have different environments; do not assume local sessions run in a cloud sandbox.",
          "source_url": "https://code.claude.com/docs/en/overview",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Account, session mode and policy dependent",
          "region_restrictions": "Not established"
        },
        {
          "name": "Default approvals: Security docs say Auto is the starting mode for interactive terminal and VS Code sessions. A classifier reviews actions instead of routine human prompts; explicit ask/deny rules still apply.",
          "source_url": "https://code.claude.com/docs/en/security",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Account, session mode and policy dependent",
          "region_restrictions": "Not established"
        },
        {
          "name": "Permission modes: Manual asks for edits and non-read-only commands; acceptEdits permits edits/common filesystem work; plan avoids source edits; dontAsk denies calls needing prompts; bypass skips most prompts and belongs only in isolated environments.",
          "source_url": "https://code.claude.com/docs/en/permissions",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Account, session mode and policy dependent",
          "region_restrictions": "Not established"
        },
        {
          "name": "Rules and extensions: Deny, ask, allow precedence is documented, but installed mods can replace some decisions depending on plan/managed policy. CLAUDE.md instructions do not enforce permissions. Review plugins, hooks and mods.",
          "source_url": "https://code.claude.com/docs/en/permissions",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Account, session mode and policy dependent",
          "region_restrictions": "Not established"
        },
        {
          "name": "MCP and custom workflows: MCP connects apps such as Drive, Jira and Slack. Instructions, skills, hooks and parallel agents extend work, but connection credentials/scopes need separate review.",
          "source_url": "https://code.claude.com/docs/en/overview",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Account, session mode and policy dependent",
          "region_restrictions": "Not established"
        },
        {
          "name": "Filesystem/network isolation: Use sandboxing for OS-level filesystem/network limits. Working-directory approval prompts alone do not stop an approved shell command from writing elsewhere under your user account.",
          "source_url": "https://code.claude.com/docs/en/security",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Account, session mode and policy dependent",
          "region_restrictions": "Not established"
        }
      ],
      "permissions": [
        "Tiered permission modes re-verified 22 October 2026: plan (read-only explore), auto (runs without routine prompts; a background classifier checks actions against your request), acceptEdits, dontAsk (auto-denies anything that would prompt) and manual. Deny/ask/allow rules apply with documented precedence.",
        "'Yes, and don't ask again' approvals save as rules to .claude/settings.local.json at the repository root and apply across future sessions and worktrees in that repo; file-edit approvals last only for the session.",
        "Feedback surveys and nonessential traffic can be disabled (CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC, CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY, DISABLE_TELEMETRY, DO_NOT_TRACK); organisations can route survey ratings through their own OpenTelemetry collector.",
        "Filesystem/network isolation beyond the permission system needs OS-level sandboxing; MCP connectors (Drive, Jira, Slack and others) follow their own scopes."
      ],
      "limitations": [
        "No code-quality, success-rate or permission-enforcement test was run. Benchmarks are not a product comparison.",
        "Old ask-before-every-edit summary only describes Manual behavior, not today's documented Auto starting mode. Surface, version, saved rules, hooks and mods matter.",
        "MCP connections and approved shell commands may disclose data or change external systems; a local tool is not an offline tool.",
        "Subscription limits are shared with other Claude use; API and third-party provider charges are separate. No fixed tasks-per-dollar guarantee.",
        "No region eligibility, numeric Max-tier allowance or exact team-seat total confirmed here. Check chosen plan and billing route."
      ],
      "runtime": [
        "Local terminal/IDE execution with model requests sent over the network",
        "Desktop and browser options; hosted cloud sessions use isolated VMs with cloned repository and outbound security proxy",
        "Provider selection, admin policy and session mode change available controls"
      ],
      "privacy_security": "Re-verified 22 October 2026 against Claude Code's data-usage doc and Anthropic's privacy policy (effective 10 September 2026). Consumer Free/Pro/Max: training happens only when the voluntary model-improvement setting is on (applies to Claude Code use from those accounts); retention is 5 years with the setting on, 30 days with it off. Commercial (Team, Enterprise, API, third-party platforms, Gov): no training on code or prompts unless the customer opts in (e.g. Developer Partner Program, first-party API only); standard retention 30 days; zero data retention available per-organisation for eligible Enterprise accounts, not by default. Local CLI transcripts are stored in plaintext under ~/.claude/projects/ for 30 days by default (cleanupPeriodDays adjustable; Desktop/Cowork sessions exempt). /feedback, /bug and /share transcripts are retained 5 years; nothing is sent until you choose. Individual cloud sessions can be deleted at any time, permanently removing their event data. Vendor documentation, not independently audited.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-28",
        "source_url": "https://code.claude.com/docs/en/overview",
        "note": "Pro $20/month when billed monthly or $200 billed annually (page shows rounded $17/month annual equivalent), with Claude Code included. Max starts at $100/month with 5x/20x usage options. Free comparison table excludes Code. Console API and third-party provider billing are separate, and plan usage limits apply; taxes excluded. Exact team seat eligibility, regional prices and per-task cost are not established here. Re-verified 22 October 2026 (claude.com/pricing now lands on code.claude.com/docs/en/overview): Pro $20 monthly or $200/year ($17/month equivalent), Max from $100/month. Re-verified 28 October 2026 (claude.com/pricing + Max plan help article 11049741 + code.claude.com/docs/en/costs): Max tiers confirmed $100/month (5x) and $200/month (20x), monthly-only; Pro annual $200/year ($17/month equivalent); Enterprise $20/seat/month plus usage at API rates, billed annually; usage credits available past plan limits at standard API rates; vendor cost guidance cites ~$13/developer/active day average and $150-250/developer/month across enterprise deployments (vendor figures)."
      },
      "sources": [
        "https://code.claude.com/docs/en/overview",
        "https://code.claude.com/docs/en/permissions",
        "https://code.claude.com/docs/en/security",
        "https://code.claude.com/docs/en/data-usage",
        "https://claude.com/pricing",
        "https://www.anthropic.com/legal/privacy"
      ],
      "changelog": [
        {
          "date": "2026-10-09",
          "change": "Re-read official features, permissions, pricing and privacy sources; added product-specific evidence table and narrow-start steps. Vendor documentation, no hands-on tests."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-10-01",
          "change": "Anthropic's v2.1.287 release notes (1 October 2026) add Claude Mods plugins, a built-in side-agent mod and MCP URL prompts. Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.287"
        },
        {
          "date": "2026-09-30",
          "change": "Release v2.1.286 (vendor release notes): permission prompts show an N-of-M count when several stack up."
        },
        {
          "date": "2026-09-29",
          "change": "Release v2.1.285 (vendor release notes): allowedProviders managed setting; option to disable WebFetch."
        },
        {
          "date": "2026-09-28",
          "change": "Release v2.1.284 (vendor release notes): Claude Sonnet 5.5 default Sonnet model on the API; vendor states 1M context, $2/$10 per Mtok."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: data-usage and permissions docs re-read - all training/retention claims verified verbatim (5y/30d consumer, commercial no-train default, ZDR per-org, plaintext local transcripts, 5-year feedback retention); added per-session cloud deletion, telemetry-disable env vars, repo-root approval-rule saving; Anthropic privacy policy (10 Sep 2026) added as source."
        },
        {
          "date": "2026-10-28",
          "change": "Field-level pricing audit: claude.com/pricing, Max plan help article and Claude Code cost docs re-read - Max $100 (5x) / $200 (20x) monthly-only confirmed, Enterprise $20/seat + API rates annual, usage credits at standard API rates, vendor cost guidance recorded. New pricing guide published; checked_label unchanged (not a full-record check)."
        }
      ],
      "coverage_rows": [
        [
          "Repository work",
          "Documented",
          "Reads a codebase, edits multiple files, runs tests/commands and documents bugs or features. Generated code still needs review and your test suite.",
          "https://code.claude.com/docs/en/overview"
        ],
        [
          "Git and CI",
          "Documented",
          "Branches, commits and pull requests; GitHub Actions and GitLab CI/CD for reviews/triage. A repository change is not automatically a safe merge.",
          "https://code.claude.com/docs/en/overview"
        ],
        [
          "Where it runs",
          "Surface-dependent",
          "Terminal, VS Code/JetBrains, desktop and web. Local CLI and Anthropic-hosted cloud sessions have different environments; do not assume local sessions run in a cloud sandbox.",
          "https://code.claude.com/docs/en/overview"
        ],
        [
          "Default approvals",
          "Current correction",
          "Security docs say Auto is the starting mode for interactive terminal and VS Code sessions. A classifier reviews actions instead of routine human prompts; explicit ask/deny rules still apply.",
          "https://code.claude.com/docs/en/security"
        ],
        [
          "Permission modes",
          "Configurable",
          "Manual asks for edits and non-read-only commands; acceptEdits permits edits/common filesystem work; plan avoids source edits; dontAsk denies calls needing prompts; bypass skips most prompts and belongs only in isolated environments.",
          "https://code.claude.com/docs/en/permissions"
        ],
        [
          "Rules and extensions",
          "Scope matters",
          "Deny, ask, allow precedence is documented, but installed mods can replace some decisions depending on plan/managed policy. CLAUDE.md instructions do not enforce permissions. Review plugins, hooks and mods.",
          "https://code.claude.com/docs/en/permissions"
        ],
        [
          "MCP and custom workflows",
          "Documented",
          "MCP connects apps such as Drive, Jira and Slack. Instructions, skills, hooks and parallel agents extend work, but connection credentials/scopes need separate review.",
          "https://code.claude.com/docs/en/overview"
        ],
        [
          "Filesystem/network isolation",
          "Optional control",
          "Use sandboxing for OS-level filesystem/network limits. Working-directory approval prompts alone do not stop an approved shell command from writing elsewhere under your user account.",
          "https://code.claude.com/docs/en/security"
        ],
        [
          "Training and storage",
          "Account-dependent",
          "Consumer training follows privacy setting: enabled data retained 5 years, disabled 30 days. Commercial code/prompts are not trained on unless opted in; standard retention 30 days. Enterprise ZDR requires eligibility and explicit enablement.",
          "https://code.claude.com/docs/en/data-usage"
        ],
        [
          "Local logs and feedback",
          "Separate risks",
          "CLI local transcripts are plaintext, normally 30 days; Desktop/Cowork continuation exceptions apply. Feedback/bug/share reports retained 5 years; optional survey transcript sharing has different 6-month terms.",
          "https://code.claude.com/docs/en/data-usage"
        ],
        [
          "Price/access",
          "Published with limits",
          "Pro $20 monthly or $200 annually includes Code; Max starts $100/month. Free plan table excludes Code. Console API/provider billing and team eligibility are separate; limits and taxes apply.",
          "https://claude.com/pricing"
        ]
      ],
      "setup_steps": [
        "Choose terminal/IDE/desktop/web from the official overview and check account eligibility before installing. Do not assume the Free chat plan includes Code.",
        "Use a version-controlled test repository or branch. Read the install route for your operating system; review the installer rather than blindly running a copied command.",
        "Inspect the actual starting mode and /permissions, including persistent repository rules. Start with Manual or Plan when you want hands-on review; use OS sandboxing for real file/network boundaries.",
        "Connect only needed MCP tools; inspect project instructions, hooks/plugins/mods before trusting a repository. Do not add production credentials to a test task.",
        "Ask for one small change, review its diff and run your own tests before committing or merging. Check privacy settings, local logs, and subscription/API cost afterwards."
      ],
      "price_tags": [
        "listed"
      ],
      "price_summary": "Pro $20/month includes Claude Code; Max $100-$200/month; API token billing option",
      "page_url": "https://agentcomparison.net/agents/claude-code/"
    },
    {
      "id": "comma",
      "name": "Comma",
      "vendor": "Comma",
      "kind": "product",
      "category": "personal assistant",
      "group": "errands-travel-food",
      "summary": "Comma describes a personal agent organised around ongoing Tasks, with connected apps, Mac access and a cloud computer.",
      "one_liner": "A personal agent that works through ongoing tasks with a cloud computer.",
      "best_fit": "Personal or team app tasks that continue in a cloud environment.",
      "poor_fit": "A user who needs every connected-app action to wait for approval without configuring boundaries.",
      "decision_note": "Distinguish cloud work from tasks needing Mac files or call recording; check the actual approval settings before connecting apps.",
      "availability": "Official documentation available; signup and plan/region details vary. See pricing/access notes.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Tasks that continue on a cloud computer",
          "source_url": "https://comma.surf/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Connected app and file work",
          "source_url": "https://comma.surf/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Routines and proactive messages",
          "source_url": "https://comma.surf/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Signal, Telegram and WeChat access",
          "source_url": "https://comma.surf/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Router/Worker model: one Router hands each Task to a Worker; choose a model per role or let Codex or Claude Code be the Worker",
          "source_url": "https://comma.surf/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Connected apps get scoped access per the data-use page.",
        "Always-on, proactive, sessionless agent: one agent keeps work moving across Tasks on a board (In progress / Needs review / Done).",
        "BYO model option: use your own API key, or your ChatGPT or Claude plan - those calls use no Comma credits.",
        "The cloud computer consumes credits while it runs; model calls are charged at the model's published API price in credits."
      ],
      "limitations": [
        "Call recording and Mac-local files need the Mac",
        "Landing-page demonstrations are not verified task outcomes"
      ],
      "runtime": [
        "Own cloud computer; some Mac-specific work needs the Mac"
      ],
      "privacy_security": "Comma's data-use page states conversations, prompts and files are never used for training and personal data is not sold; user-submitted content is not used to train its own or third-party general-purpose AI models, except where a customer expressly requests a separate training or fine-tuning workflow under specific terms. Connected apps get scoped access. A published retention table covers the main categories: workspace content and saved workflow context retained until deleted; intermediate outputs typically deleted within 30 days unless promoted to persistent state; recordings and uploaded audio retained until deleted. Checked 23 October 2026.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-22",
        "source_url": "https://comma.surf/",
        "note": "Value $0/month. Pricing toggle checked 7 October shows Pro $80/month and $800/year with Save $160; Billed yearly label persists in both states, so confirm final checkout period. About 300-600 tasks/month is a vendor estimate. 1M credits = US$1; cloud computer about 74,000/hour while running; model calls at API cost. BYOK model calls use no Comma credits, but compute still does. Credits do not roll over. Re-checked 22 October 2026 on comma.surf (the /pricing path now redirects to the homepage pricing section): Free $0 BYOK and Pro $80/month billed yearly (save $160) unchanged. Monthly Pro $80 re-verified 22 October 2026; the annual $800 figure is from the 7 October toggle read and was not re-verifiable in the 22 October text pass. New from the 22 October FAQ: 1M credits = US$1 of usage at published API prices; the cloud computer uses about 74,000 credits per running hour, billed by the second and shut down ~5 minutes after last work; plan credits expire monthly."
      },
      "sources": [
        "https://comma.surf/",
        "https://comma.surf/pricing",
        "https://comma.surf/docs",
        "https://comma.surf/data-use"
      ],
      "changelog": [
        {
          "date": "2026-10-23",
          "change": "Field-level audit: pricing re-confirmed from verified record (free $0; Pro $80/mo or $800/yr annual); data-use page captured - never trains on conversations/prompts/files, no general-purpose model training on user content, no data sale; credit mechanics (API-price passthrough, monthly expiry, BYO key/plan free of credits, cloud computer consumes); retention table. Coverage rows, permissions, setup steps, privacy fields filled. No hands-on testing."
        },
        {
          "date": "2026-10-07",
          "change": "Rechecked vendor plans, pricing and policies; clarified payment/training scope and unresolved details."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-10-22",
          "change": "Homepage and pricing re-read: Free $0 and Pro $80/month re-verified; credit economics documented (1M credits = $1; cloud computer ~74k credits/hour, billed per second); Router/Worker model added; WeChat no longer visible on the homepage (kept in channels per earlier docs, softened on the comparison page); annual $800 left dated 7 October."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Free $0; Pro $80/mo or $800/yr",
          "Record verified 7-22 Oct: free tier $0/month; Pro $80/month or $800/year billed annually ('Save $160'). Monthly credits expire after one month; credit packs last to end of the purchase month.",
          "https://comma.surf/pricing"
        ],
        [
          "Training use",
          "Never, per data-use page",
          "Conversations, prompts and files never used for training; no training of own or third-party general-purpose models on user content; exception only for express customer-requested training workflows.",
          "https://comma.surf/data-use"
        ],
        [
          "Credit model",
          "API-price passthrough",
          "Model calls cost the model's published API price in credits (e.g. a $2/M-token model uses 2 credits per input token); BYO key/ChatGPT/Claude plan uses no credits; cloud computer consumes credits while running.",
          "https://comma.surf/pricing"
        ],
        [
          "Retention",
          "Published table",
          "Workspace content until deleted; intermediate outputs typically 30 days; recordings/audio until deleted; configurable.",
          "https://comma.surf/data-use"
        ]
      ],
      "setup_steps": [
        "Start on the free tier and run one Task end-to-end (In progress to Done) to see the board model before paying.",
        "Check Settings > Usage & billing for the credit renewal date - monthly credits expire after one month.",
        "If you already pay for ChatGPT or Claude, connect your own key or plan so model calls skip Comma credits."
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Value $0; Pro toggle $80/month or $800/year; confirm checkout period",
      "page_url": "https://agentcomparison.net/agents/comma/"
    },
    {
      "id": "devin",
      "name": "Devin",
      "vendor": "Cognition",
      "kind": "product",
      "category": "software engineering agent",
      "group": "coding-agents",
      "summary": "Cognition documents Devin as an autonomous software engineer that writes, runs and tests code, available as Devin Cloud sessions with an IDE, shell and browser, plus Devin Desktop and a Devin CLI that can also steer cloud sessions from a terminal.",
      "one_liner": "An autonomous software engineer with its own IDE, shell and browser.",
      "best_fit": "Engineering teams delegating well-scoped, verifiable tickets who will review the draft PRs it returns.",
      "poor_fit": "Hard-to-verify projects with no completion criteria, or anyone needing a numeric published usage quota or default-off model training without a paid-plan opt-out.",
      "decision_note": "Keep the surfaces separate: Devin Cloud runs on Cognition-managed VMs, while the CLI runs on your own machine under configurable approval modes. Cognition says customer data may be used for model training by default, so a paid plan's Data Controls opt-out matters before connecting private repositories.",
      "availability": "Free, Pro, Max and Teams plans sign up at app.devin.ai; Enterprise is sold directly. Devin Desktop and the CLI download separately; the pricing page lists Devin Cloud access from Pro up.",
      "last_checked": "2026-10-11",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Writes, runs and tests code in a cloud workspace",
          "source_url": "https://docs.devin.ai/get-started/devin-intro",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Devin Cloud listed from Pro up on the pricing page",
          "region_restrictions": null
        },
        {
          "name": "Embedded IDE, shell and interactive browser, with takeover",
          "source_url": "https://docs.devin.ai/get-started/devin-intro",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Devin CLI with tiered approval modes and optional OS sandbox",
          "source_url": "https://docs.devin.ai/cli/reference/permissions",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Smart mode rolling out gradually",
          "region_restrictions": null
        },
        {
          "name": "Start, steer and resume Devin Cloud sessions from the terminal",
          "source_url": "https://docs.devin.ai/cli/cloud",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Admin security profiles restricting network, MCP, git and GitHub CLI access",
          "source_url": "https://docs.devin.ai/product-guides/security-profiles",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Enterprise adds cross-organisation mandatory profiles",
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Documentation shows IDE takeover and an interactive browser for cloud sessions",
        "In the CLI's default Normal mode, read-only actions auto-approve while file writes and shell commands prompt; Smart mode auto-runs only actions a fast model judges safe and never auto-approves installs, rm, sudo, mutating git or access to dotenv and key material; Bypass auto-approves everything; Autonomous mode pairs with an OS sandbox where shell and fetches auto-approve but edit/write tools still prompt",
        "Organisation-level deny and ask rules override every user permission mode",
        "Admin security profiles can allowlist network destinations and MCP servers, set git read-only or full access, remove the GitHub CLI token and make Devin's own management tools read-only; sessions without a profile run unrestricted",
        "Secrets can be scoped to the organisation, a person, a repository or one session, are encrypted at rest and are bound only to the commands that need them"
      ],
      "limitations": [
        "Teams pricing conflicts: pricing page says $80/month plus $40 per full seat; billing docs say an $80 minimum met by seats and credits. Confirm the offered total before payment.",
        "Vendor explicitly says documentation may be out of date",
        "Quotas are described as daily and weekly allowances but not published as numbers",
        "Model training is on by default; only paid plans can opt out, and on Teams only an administrator can",
        "Security profiles are admin-configured: without one, sessions run unrestricted, and on customer-run Outposts the network policy must be enforced by the operator's own infrastructure",
        "Smart approval mode is rolling out gradually and may not be available"
      ],
      "runtime": [
        "Devin Cloud: sessions on Cognition-managed VMs with shell, IDE and interactive browser, driven from the web app, Desktop, API or devin --cloud",
        "Devin CLI on your own machine, with Normal, Accept Edits, Smart, Bypass and sandboxed Autonomous permission modes"
      ],
      "privacy_security": "Re-verified 22 October 2026 against Cognition's Platform Terms of Service (updated 30 June 2026) and privacy policy (updated 9 March 2026). Customer Data may be used for model training and service improvement by DEFAULT (ToS 3.3.1). Paid tiers can opt out; after opt-out, Customer Data is not used for training and zero data retention is enabled with Cognition's model providers. For the Teams tier only an administrator may opt out; Enterprise customer data is never used for training without express prior written consent (per Cognition's security page). ZDR does not preclude retention of data flagged by automated safety/abuse classifiers, safety/AUP review, or legal process (ToS 3.3.3). The privacy policy adds: voice-input audio is deleted after transcription unless stated otherwise, and deletion of personal information can be requested. Use restrictions prohibit building competing products or training competing models. Vendor documents, not independently audited.",
      "pricing": {
        "amount": 0,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Free $0; Pro $20/month; Max $200/month; Teams $80/month minimum plus $40/month per full seat",
        "usage_charges": "On-demand credits at API pricing past quota; an eligible ChatGPT Go, Plus or Pro subscription can pay for GPT model usage so it does not count against the Devin quota",
        "verified_at": "2026-10-11",
        "source_url": "https://devin.ai/pricing",
        "note": "The pricing page lists Free at $0 with a light quota, limited models, unlimited inline edits and Tab completions; Pro at $20/month adding full model availability, Devin Cloud access, free SWE-2 use in Desktop and CLI through 16 October 2026, and extra usage at API pricing; Max at $200/month with significantly higher quotas; Teams at an $80/month minimum plus $40/month per full seat for up to 200 users, with free flex seats that draw shared on-demand credits. Billing docs describe daily and weekly quotas for Pro and full seats (weekly only on Max) but publish no numbers. Reviewing a public GitHub PR at devinreview.com is free without an account. Teams source conflict: the billing guide instead describes an $80 minimum met by full seats and credits. Confirm the offered total; these two models are not interchangeable."
      },
      "sources": [
        "https://devin.ai/pricing",
        "https://docs.devin.ai/get-started/devin-intro",
        "https://docs.devin.ai/admin/billing/self-serve",
        "https://docs.devin.ai/cli/reference/permissions",
        "https://docs.devin.ai/cli/cloud",
        "https://docs.devin.ai/product-guides/security-profiles",
        "https://docs.devin.ai/admin/security",
        "https://cognition.com/legal/privacy-policy",
        "https://cognition.com/legal/platform-terms-of-service"
      ],
      "changelog": [
        {
          "date": "2026-10-11",
          "change": "Expanded product-specific feature evidence and narrow setup from fresh official docs. Costs, credential exposure and limits remain explicit. Documentation research, not tested."
        },
        {
          "date": "2026-10-10",
          "change": "Flagged conflicting official Teams billing models and corrected pricing guide membership cap to 200. Pricing sources reread; no other product fields reverified."
        },
        {
          "date": "2026-10-08",
          "change": "Refreshed from the pricing page and seven official docs: added the default-on model-training opt-out and zero-data-retention claim, retention scope, SOC 2 Type II statement, admin security profiles, CLI approval modes and secrets scoping. Pricing synced with devin.ai/pricing. Documentation research only; not tested."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-09-29",
          "change": "Cognition blog: GPT-6.1 Sol available in Devin Desktop and Devin CLI. Benchmark figures in the post are Cognition's own."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: training-default-on + paid opt-out + ZDR claims re-verified against the current Platform ToS (30 June 2026) including the safety/abuse and legal-process exceptions to ZDR; privacy policy (9 March 2026) adds voice-audio deletion after transcription; legal URLs corrected to cognition.com (devin.ai/terms and /privacy 404)."
        }
      ],
      "coverage_rows": [
        [
          "Cloud versus local",
          "Separate controls",
          "Cloud uses managed VMs. Local CLI permissions must not be treated as proof of Cloud behaviour.",
          "https://docs.devin.ai/get-started/devin-intro"
        ],
        [
          "CLI Normal and Accept Edits",
          "Documented",
          "Normal auto-approves reads, prompts for writes/shell. Accept Edits auto-approves workspace edits, but shell/outside writes still prompt.",
          "https://docs.devin.ai/cli/reference/permissions"
        ],
        [
          "Smart and Bypass",
          "Different risk",
          "Smart judges routine actions and prompts on uncertainty; sensitive categories never auto-approve. Bypass auto-approves all tool calls; organisation deny/ask rules still apply.",
          "https://docs.devin.ai/cli/reference/permissions"
        ],
        [
          "Autonomous sandbox",
          "Not universal write permission",
          "Sandbox allows shell/fetch automatically within OS limits. Direct edit/write tools still prompt and granted write scope can expand the sandbox.",
          "https://docs.devin.ai/cli/reference/permissions"
        ],
        [
          "Cloud security profiles",
          "Must be configured",
          "Profiles can restrict network, MCP, git and GitHub CLI access. No-profile sessions run unrestricted; customer Outpost network controls depend on operator infrastructure.",
          "https://docs.devin.ai/product-guides/security-profiles"
        ],
        [
          "Training and retention",
          "Plan-dependent vendor claim",
          "Training may be on by default. Paid users can opt out; Teams requires an admin. Enterprise training needs express written consent. Opt-out enables provider zero-retention, not deletion of Cognition-held records.",
          "https://docs.devin.ai/admin/security"
        ],
        [
          "Teams pricing",
          "Official source conflict",
          "Pricing page says $80 base plus $40/full seat; billing docs describe an $80 minimum met by seats/credits. Both cap self-serve Teams at 200. Confirm the actual offer.",
          "https://docs.devin.ai/admin/billing/self-serve"
        ],
        [
          "Quotas and credits",
          "Numeric quotas unknown",
          "Pro/full seats have daily/weekly allowance; Max weekly only. Teams automations/review use shared credits, not seat quota. Numeric allowances are not published here.",
          "https://docs.devin.ai/admin/billing/self-serve"
        ]
      ],
      "setup_steps": [
        "Choose Cloud or local CLI first, then one small ticket with tests and a reviewable diff.",
        "Before private repository access, review training opt-out and organisation permission rules.",
        "Use Normal mode for local review or an explicitly configured Cloud security profile; do not select Bypass as a safety default.",
        "Confirm quota, shared credits and the actual Teams quote before paid work."
      ],
      "price_tags": [
        "free",
        "listed",
        "usage"
      ],
      "price_summary": "Free $0; Pro $20/month; Max $200/month; Teams $80/month minimum plus $40/month per full seat",
      "page_url": "https://agentcomparison.net/agents/devin/"
    },
    {
      "id": "fin",
      "name": "Fin",
      "vendor": "Intercom",
      "kind": "product",
      "category": "customer-facing service and sales agent",
      "group": "work-business",
      "summary": "Intercom positions Fin as a customer agent across service and sales, combining skills and knowledge with customer history and context.",
      "one_liner": "Intercom's customer service and sales agent, billed per outcome.",
      "best_fit": "Customer-facing service and sales workflows.",
      "poor_fit": "Choosing a vendor solely from its performance superlatives.",
      "decision_note": "Vendor superlatives are not verified performance evidence. Compare escalation quality and unsafe-answer rate on a fixed test set.",
      "availability": "Official documentation available; signup and plan/region details vary. See pricing/access notes.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Service and sales workflows",
          "source_url": "https://www.intercom.com/fin",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Customer knowledge and context",
          "source_url": "https://www.intercom.com/fin",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Rules and guidance can override default escalation; a teammate/inbox routing target is needed for human handover.",
        "Connector authentication and Data Hub identity verification need configuration. Email Ask for input can send its original draft after a timeout, not wait indefinitely for approval."
      ],
      "limitations": [
        "Email teammate input is open beta and unavailable on standalone Fin plans.",
        "Manual connector triggering is workflow control, not an automatic human approval requirement.",
        "Verify your outcome definitions and contract; full sales qualification is not the $0.99 self-serve outcome."
      ],
      "runtime": [
        "Vendor-managed customer-agent offering"
      ],
      "privacy_security": "Verified 22 October 2026 across three current documents. (1) fin.ai/training: the Fin AI Group is beginning in-house training of custom models on successfully resolved conversations, with identifying information removed first via Microsoft Presidio; fine-tuned models are used only to improve Fin, under its ISO 42001 governance, and customers can opt out at any time via a help-article process. (2) Master Subscription Agreement (effective 8 September 2026) 3.2: customers grant Intercom the right to use Customer Data 'to provide and improve our Services, including for Intercom AI model training'; 3.4 lets Intercom de-identify Customer Data into Anonymized Data it owns and may use during and after the term. (3) Privacy policy (effective 21 October 2026): publicly available online information is used for AI model training with steps to remove personal information first. Earlier specifics still per vendor but not re-verified today: Fin fine-tuning deletion within 30 days, third-party AI providers barred from training with zero retention, and connector response logs of 14 days (7 on request). Vendor policy, not independently audited.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-22",
        "source_url": "https://www.intercom.com/pricing",
        "note": "Intercom lists Fin from USD $0.99 per outcome. Seat fees and a minimum monthly commitment may also apply. Outcome definitions, eligible support channels and your contract affect the bill; do not treat $0.99 as the total monthly cost."
      },
      "sources": [
        "https://fin.ai/",
        "https://www.intercom.com/pricing",
        "https://www.intercom.com/help/en/articles/12396892-manage-fin-ai-agent-s-escalation-guidance-and-rules",
        "https://www.intercom.com/help/en/articles/9916497-how-to-set-up-data-connectors",
        "https://www.intercom.com/help/en/articles/16340024-using-the-data-hub-to-personalize-fin-s-responses",
        "https://www.intercom.com/help/en/articles/8205718-fin-ai-agent-outcomes",
        "https://www.intercom.com/security",
        "https://fin.ai/training",
        "https://www.intercom.com/legal/terms-and-policies",
        "https://www.intercom.com/legal/privacy"
      ],
      "changelog": [
        {
          "date": "2026-10-14",
          "change": "Expanded product-specific runtime, approval, access and privacy evidence with narrow setup steps. Documentation research, not hands-on testing."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: in-house model training program (Presidio anonymization, ISO 42001, opt-out) verified at fin.ai/training; MSA 3.2/3.4 training and anonymized-data rights and the 21 October 2026 privacy policy added; 30-day deletion and connector-log retention left dated pending re-verification of the security page."
        }
      ],
      "coverage_rows": [
        [
          "Escalation behavior",
          "Configurable, not guaranteed",
          "Rules and natural-language guidance can override default human-request, frustration and loop handling, including asking more questions or preventing escalation. Workflows handle routing after escalation.",
          "https://www.intercom.com/help/en/articles/12396892-manage-fin-ai-agent-s-escalation-guidance-and-rules"
        ],
        [
          "Human routing",
          "Requires a target",
          "Bot-only workflows without a teammate or inbox target do not offer human escalation. Fin uses conversational handover text; guidance does not control quick-reply buttons.",
          "https://www.intercom.com/help/en/articles/12396892-manage-fin-ai-agent-s-escalation-guidance-and-rules"
        ],
        [
          "Email teammate input",
          "Beta with a timeout",
          "Ask for input is email-only open beta, unavailable on standalone Fin plans. A teammate can guide the reply, approve the original draft or take over. Proceed after automatically sends the original draft if nobody responds; this is not a mandatory approval gate.",
          "https://www.intercom.com/help/en/articles/12396892-manage-fin-ai-agent-s-escalation-guidance-and-rules"
        ],
        [
          "Connector triggers",
          "Automatic or workflow-controlled",
          "Direct trigger lets Fin call a connector from a customer question. Manual trigger requires a Workflow, Procedure or Macro, recommended for sensitive or write-active operations. Manual does not itself mean a human approves every action.",
          "https://www.intercom.com/help/en/articles/9916497-how-to-set-up-data-connectors"
        ],
        [
          "Customer authentication",
          "Explicit configuration",
          "A connector authentication toggle enforces workspace rules. Data Hub attribute access requires identity verification, one conversation participant and supported Fin for Service channels: Messenger, email or API.",
          "https://www.intercom.com/help/en/articles/16340024-using-the-data-hub-to-personalize-fin-s-responses"
        ],
        [
          "Outcome billing",
          "Five definitions",
          "Chat/email outcomes in Intercom are $0.99 for resolution, Procedure handoff, disqualification and explicitly marked Self-serve routing; full qualification is $9.99. At most one outcome per conversation. Existing outcomes are not automatically repriced.",
          "https://www.intercom.com/help/en/articles/8205718-fin-ai-agent-outcomes"
        ],
        [
          "Model training",
          "First- and third-party differ",
          "Intercom says Fin can use anonymized customer data for fine-tuning, with opt-out and deletion within 30 days. Third-party AI providers are barred from training on the data and retain none after output. This is not a blanket statement that Intercom stores nothing.",
          "https://www.intercom.com/security"
        ],
        [
          "Connector response logs",
          "A separate retention period",
          "Responses from Fin-triggered Data connectors are stored for up to 14 days; Support can provide 7-day retention on request. Customer records and other Intercom data have separate terms. No all-product retention period is established here.",
          "https://www.intercom.com/help/en/articles/9916497-how-to-set-up-data-connectors"
        ]
      ],
      "setup_steps": [
        "Start with one low-risk support intent and a small, checked knowledge source.",
        "Create a real teammate/inbox route, then test a human request, a loop and any override guidance before deployment.",
        "For sensitive connectors, use a workflow-controlled trigger and authentication; check the live configuration after publishing.",
        "If enabling email Ask for input, review who is asked and the Proceed after auto-send timeout. Compare observed outcomes with the billing guide."
      ],
      "price_tags": [
        "usage"
      ],
      "price_summary": "From $0.99 per outcome; seat fees and minimums may also apply",
      "pricing_guide": {
        "href": "/guides/fin-outcome-pricing-and-escalation/",
        "label": "Fin outcome pricing and escalation explained"
      },
      "page_url": "https://agentcomparison.net/agents/fin/"
    },
    {
      "id": "fo-wajo",
      "name": "Fo by Wajo",
      "vendor": "Wajo",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Wajo describes Fo as an action agent for real-world bookings, coordination, purchases and life administration, with human help when AI cannot complete a task.",
      "one_liner": "An action agent for bookings, purchases and life admin, with human help as backup.",
      "best_fit": "Real-world coordination and life administration with vendor-assisted escalation.",
      "poor_fit": "A user requiring verified zero-cost transactions or fully AI-only execution.",
      "decision_note": "Set spending limits explicitly; separate marketing privacy claims from the broader terms on content training, credentials and payments.",
      "availability": "Vendor help FAQ (read 27 Oct 2026) documents Fo mini (free, daily limit) and OG Fo (free, unlimited tasks; invite-gated via a social post sent to Fo or a friend's invite link). Enterprise via sales; Fo Pro announced as coming soon. Actual invitation eligibility, supported countries and individual feature access were not tested.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Real-world tasks: Scheduling, vendors, gifting, group orders and purchases. Homepage examples are marketing use cases, not observed successful outcomes.",
          "source_url": "https://wajo.ai/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Plan and permissions dependent",
          "region_restrictions": "Not established"
        },
        {
          "name": "Messaging and group work: vendor help FAQ (read 27 Oct) lists web, iMessage, SMS, RCS, Telegram, Slack for workspaces and phone calls as channels; there is no app, and WhatsApp, Signal and Messenger are unavailable. Group texts can include up to 8 people not on Wajo. Exact group-to-private disclosure rules are not established.",
          "source_url": "https://wajo.ai/help",
          "observed_at": "2026-10-27",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Plan and permissions dependent",
          "region_restrictions": "Fo's numbers are US numbers; SMS group texts work for US/Canada numbers only"
        },
        {
          "name": "Connected tools: Homepage FAQ names Gmail, Calendar, Drive, Slack and permissioned connections; custom agents may connect CRM/storage/forms/internal tools. Availability differs by agent and plan.",
          "source_url": "https://wajo.ai/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Plan and permissions dependent",
          "region_restrictions": "Not established"
        },
        {
          "name": "Proactive/shared agents: Defined work can run independently; business teams can share agents/knowledge/workflows with role-based instructions and approvals. No universal reminder cadence or reliability guarantee.",
          "source_url": "https://wajo.ai/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Plan and permissions dependent",
          "region_restrictions": "Not established"
        },
        {
          "name": "Human backup: Trained executive assistants can help when AI cannot finish. Verify what non-Google data those helpers see and how approval/escalation is configured.",
          "source_url": "https://wajo.ai/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Plan and permissions dependent",
          "region_restrictions": "Not established"
        },
        {
          "name": "MCP server and API: a hosted MCP endpoint (mcp.api.wajo.ai/mcp) lets MCP-capable clients (Claude Code, Claude.ai and Claude Desktop, Cursor, ChatGPT, VS Code) build, manage and chat with Wajo agents end-to-end; a separate /fo profile (ask_fo) lets consumer apps host Fo. Auth via API key (wak_ prefix) or OAuth. Signed webhooks fire on conversation turn close, scheduled actions and proactive messages.",
          "source_url": "https://wajo.ai/docs/mcp-server.md",
          "observed_at": "2026-10-27",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Account required; all tools are owner-scoped",
          "region_restrictions": "Not established"
        }
      ],
      "permissions": [
        "Calls, emails and coordinates with people to complete real-world tasks (bookings, quotes, cancellations, group orders), including group chats.",
        "Works with trained executive assistants when a task needs a human - human help is part of the service model.",
        "Pays with single-use cards per purchase; emails from its own address, never as you; signs in to websites without seeing or storing passwords (vendor-stated).",
        "Can work in draft-only mode, ask for approval before acting, or handle defined work independently.",
        "On calls to anyone other than you, Fo introduces itself as your assistant, states that the call is recorded and says it is an AI assistant if asked; the recording notice cannot be turned off (help FAQ, read 27 Oct 2026)."
      ],
      "limitations": [
        "Free plan access does not cover merchant purchases or establish zero service fees.",
        "Homepage credential protection and payment-card slogans are narrower than all routes described in the terms.",
        "No comparative reliability testing was run.",
        "Recurring-check controls, contacts-import setup and card-link expiry remain unestablished. Physical-device control (lamp reminders, vacuum, phone callbacks) announced in the vendor's 19 October launch posts is built on webhooks and MCP but is not yet described in vendor documentation and was not tested."
      ],
      "runtime": [
        "Vendor-hosted action agent through email,Slack,text,phone and web",
        "Personal group coordination and business shared agents depend on plan/permissions",
        "Human assistant escalation can be part of execution"
      ],
      "privacy_security": "Wajo's homepage and privacy policy (re-read 23 October 2026) state: your data is never used to train third-party AI models; Google API data is not used to develop, improve or train generalised or non-personalised AI/ML models; personal information is not sold; OAuth tokens and cached Google data are encrypted in transit and at rest. Purchases use single-use cards so real card details are never shared with merchants or seen by Fo; a payment method is saved only with approval and is revocable. Fo emails from its own address, never as you, and signs in to websites without seeing or storing passwords (vendor-stated mechanism). Source content may be temporarily retained in session records needed to provide requested features. Human executive assistants may be brought into tasks.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-27",
        "source_url": "https://wajo.ai/help",
        "note": "Re-checked 27 October 2026 via the vendor help FAQ (wajo.ai/help): OG Fo is free with unlimited tasks, invite-gated (post about Fo and send Fo the link, or use a friend's invite link; Wajo says it plans to keep OG Fo free). Fo mini is free with a daily limit. Enterprise (Slack, teams, custom voice and tagline) is priced per setup via sales@wajo.ai. Fo Pro is 'coming soon' with no date or price. No public price list exists. Purchases use single-use cards; possible service fees per the terms. A 19 October vendor launch post offered access code FOMCP for OG Fo (not verified).",
        "class": "free"
      },
      "sources": [
        "https://wajo.ai/",
        "https://wajo.ai/terms-of-service",
        "https://wajo.ai/privacy",
        "https://wajo.ai/help",
        "https://wajo.ai/docs/mcp-server.md",
        "https://wajo.ai/docs/mcp-tools.md",
        "https://x.com/shivanipod/status/2112233722846355593"
      ],
      "changelog": [
        {
          "date": "2026-10-09",
          "change": "Re-read official features, permissions, pricing and privacy sources; added product-specific evidence table and narrow-start steps. Vendor documentation, no hands-on tests."
        },
        {
          "date": "2026-10-07",
          "change": "Rechecked vendor plans, pricing and policies; clarified payment/training scope and unresolved details."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-10-22",
          "change": "Homepage re-read: Fo Mini free (1,000 tasks/day) and OG Fo free (unlimited) with invite to lift limits re-verified; draft-only/ask-before-acting/independent modes and ask-before-hard-to-reverse-actions confirmed verbatim on current page."
        },
        {
          "date": "2026-10-23",
          "note": "Fresh source-to-field re-check (phase 2): homepage restructured; Fo Mini/OG Fo pricing removed (cost now unconfirmed); new documented privacy features added (single-use payment cards, own-address email, credential-less sign-in, no third-party training); human executive-assistant escalation confirmed. Counts toward the detailed ledger again."
        },
        {
          "date": "2026-10-26",
          "change": "CTR polish (non-test, near-page-1 work): meta title/description now answer the pricing query with the documented 23 Oct status (price list removed, cost unconfirmed). checked_label unchanged."
        },
        {
          "date": "2026-10-27",
          "change": "Field-level audit supporting near-page-1 wajo queries (avg pos 6.0): help FAQ and MCP docs read. Pricing note now reflects documented free plans (Fo mini daily limit; OG Fo free unlimited, invite-gated; Enterprise by quote; Pro 'coming soon'), superseding the 23 Oct 'cost unconfirmed' note; still no public price list, so the v167 non-test title stays. Consumer channels established (iMessage/SMS/RCS/Telegram/Slack/web/phone; no app; WhatsApp/Signal/Messenger unavailable; group texts capped at 8 non-Wajo contacts; US numbers). Added hosted MCP server capability (mcp.api.wajo.ai/mcp, /fo profile, API key or OAuth) and the call AI-disclosure/recording permission. Physical-device control from the 19 Oct vendor X announcement (read via teapawt mirror; canonical URL in sources) recorded as announced-not-documented. Meta description updated to match the documented plans. checked_label unchanged (23 October 2026): homepage, terms and privacy not re-read today."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "No longer published",
          "The Fo Mini (1,000 tasks/day free) and OG Fo blocks were removed from wajo.ai between 9 and 23 October 2026; no public price list remains. Current cost is unconfirmed.",
          "https://wajo.ai/"
        ],
        [
          "Payment cards",
          "Single-use per purchase",
          "Fo checks out with a one-time card for each purchase; the real card details are never shared with the merchant or seen by Fo. A payment method is saved only with your approval and can be revoked in account settings.",
          "https://wajo.ai/"
        ],
        [
          "Email identity",
          "Sends as itself",
          "Fo emails from its own address and never sends as you.",
          "https://wajo.ai/"
        ],
        [
          "Credential handling",
          "Not seen or stored",
          "Fo signs in to websites for you without seeing or storing your password (vendor-stated mechanism).",
          "https://wajo.ai/"
        ],
        [
          "Real-world tasks",
          "Documented",
          "Scheduling, vendors, gifting, group orders and purchases. Homepage examples are marketing use cases, not observed successful outcomes.",
          "https://wajo.ai/"
        ],
        [
          "Messaging and group work",
          "Documented",
          "Email, Slack, text, phone and group coordination are listed. OG Fo lists group chats. Exact group-to-private disclosure rules and supported consumer message apps are not established.",
          "https://wajo.ai/"
        ],
        [
          "Connected tools",
          "Plan-dependent",
          "Homepage FAQ names Gmail, Calendar, Drive, Slack and permissioned connections; custom agents may connect CRM/storage/forms/internal tools. Availability differs by agent and plan.",
          "https://wajo.ai/"
        ],
        [
          "Proactive/shared agents",
          "Documented",
          "Defined work can run independently; business teams can share agents/knowledge/workflows with role-based instructions and approvals. No universal reminder cadence or reliability guarantee.",
          "https://wajo.ai/"
        ],
        [
          "Human backup",
          "Documented, access gap",
          "Trained executive assistants can help when AI cannot finish. Verify what non-Google data those helpers see and how approval/escalation is configured.",
          "https://wajo.ai/"
        ],
        [
          "Spending/approval",
          "Configure explicitly",
          "Terms 5.3 permits scoped future/repeated actions, relies on controls in account settings and may act under general authorization without configured limits. Not an approval before every purchase guarantee.",
          "https://wajo.ai/terms-of-service"
        ],
        [
          "Cards and saved accounts",
          "Two routes",
          "Homepage describes single-use cards. Terms also allow purchases using payment methods saved with merchants and secure authorized credential storage. Do not assume every route uses one-time cards.",
          "https://wajo.ai/terms-of-service"
        ],
        [
          "Call recording",
          "Consent required",
          "Terms 5.9-5.11 describes recording/transcription, notice at the beginning and the user's responsibility for required consents; marketing/prerecorded-call restrictions apply.",
          "https://wajo.ai/terms-of-service"
        ],
        [
          "Training privacy",
          "Scope distinction",
          "Homepage excludes third-party training. Terms 6.3 permits de-identified/aggregated training for Wajo services. Google Workspace API data has separate generalized-training restrictions.",
          "https://wajo.ai/privacy"
        ],
        [
          "Retention/deletion",
          "No fixed window",
          "Purpose-based retention can include derived context; deleting content may leave copies. Revocation is different from erasing previously stored data.",
          "https://wajo.ai/privacy"
        ],
        [
          "Plans",
          "Published access, gaps",
          "Mini free 1,000 tasks/day; OG free unlimited tasks/call time/group chats; Pro contact vendor with heavier workload, MCP/custom agents. Invitation eligibility and service fees remain unknown.",
          "https://wajo.ai/"
        ]
      ],
      "setup_steps": [
        "Read the homepage's current signup route and chosen plan. Free access is not free merchant purchases; confirm invite eligibility and possible service fees.",
        "Define a single low-stakes task and whether Fo should draft only, ask before acting, or act within narrow standing limits.",
        "Before enabling purchases, configure account spending limits and exact scope. Terms allow repeated/future authorization, so do not rely on a generic ask-before-acting slogan.",
        "Connect only necessary services. Check whether a human helper may receive task data and get required contact/call-recording consent.",
        "Review the booking, recipient, total and result; check any saved merchant card or credential access and revoke connections you no longer need. Request deletion separately."
      ],
      "price_tags": [
        "free"
      ],
      "price_summary": "Free plans advertised; purchases and possible service fees are separate",
      "meta_title": "Fo by Wajo pricing (Oct 2026): no public price list",
      "page_url": "https://agentcomparison.net/agents/fo-wajo/"
    },
    {
      "id": "instinct",
      "name": "Instinct",
      "vendor": "Instinct",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Instinct describes a personal assistant reached by text or call, connected to applications and devices.",
      "one_liner": "A personal assistant you text or call for email, calendar and everyday tasks.",
      "best_fit": "A messaging-first assistant connected to personal apps and devices.",
      "poor_fit": "Connecting sensitive services without checking controls and terms.",
      "decision_note": "Check permission scope and retention before connecting sensitive accounts.",
      "availability": "Official documentation available; signup and plan/region details vary. See pricing/access notes.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Text or call interface",
          "source_url": "https://instinct.com/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Connections to applications and devices",
          "source_url": "https://instinct.com/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Connects to applications and services the user grants (including Google Workspace); access is scoped by the permissions granted.",
        "Google Workspace content (emails, events, files) is covered by a no-training, no-ads commitment in the policy.",
        "Data deletion is available by asking the agent or disconnecting in settings.",
        "Acts across messaging channels (iMessage/WhatsApp/SMS per product pages) with agent-to-agent 'trusted network' coordination."
      ],
      "limitations": [
        "Disconnecting services does not automatically delete previously indexed data.",
        "No independent safety or reliability test was run."
      ],
      "runtime": [
        "Vendor-hosted personal assistant"
      ],
      "privacy_security": "Instinct's privacy policy states information may be used to evaluate, fine-tune and train the AI models powering its products, with an opt-out that applies on a go-forward basis (models previously trained before opt-out may still be used); information flagged for safety review may still be used to improve safety systems. Google Workspace data is explicitly NOT used to evaluate, fine-tune, train or improve AI models, nor for ads. The policy states Instinct does not sell user information or disclose it to registered data brokers. Users can delete data via conversation with the agent or settings disconnect. Checked 23 October 2026.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-21",
        "source_url": "https://instinct.com/terms",
        "note": "Public terms describe paid features and fees in USD but do not provide a universal price. Current signup eligibility and your offered price must be confirmed. Unknown pricing is not free.",
        "class": "see_vendor"
      },
      "sources": [
        "https://instinct.com/",
        "https://instinct.com/terms",
        "https://instinct.com/privacy-policy"
      ],
      "changelog": [
        {
          "date": "2026-10-23",
          "change": "Field-level audit from live privacy policy/terms: training use is opt-out (go-forward; safety-review exception; models trained pre-opt-out persist); Google Workspace data excluded from training and ads; no data sale or data-broker disclosure; self-serve deletion; pricing not publicly listed (terms describe USD fees). Coverage rows, permissions, setup steps, privacy fields filled. No hands-on testing."
        },
        {
          "date": "2026-10-07",
          "change": "Rechecked public terms and privacy policy; clarified the Google Workspace API training exclusion."
        },
        {
          "date": "2026-10-06",
          "change": "Added Poke vs Instinct comparison; re-read Poke pricing, docs and privacy notice and Instinct terms. No change to the record above."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-10-21",
          "change": "Terms re-read: still no universal price published. Table label corrected from Free to See vendor (unknown pricing is not free)."
        }
      ],
      "coverage_rows": [
        [
          "Training use",
          "Opt-out model",
          "Policy: information may be used to evaluate, fine-tune and train its AI models; opt-out available, go-forward only; safety-review content may still be used for safety improvements; Google Workspace data excluded from training.",
          "https://instinct.com/privacy-policy"
        ],
        [
          "Data sale",
          "Not sold",
          "Policy states information is not sold and not disclosed to registered data brokers.",
          "https://instinct.com/privacy-policy"
        ],
        [
          "Pricing",
          "Not publicly listed",
          "Terms describe paid features and fees in USD without a universal published price; current signup eligibility varies.",
          "https://instinct.com/terms"
        ],
        [
          "Deletion",
          "Self-serve",
          "Users can delete via the agent conversation or by disconnecting in settings.",
          "https://instinct.com/privacy-policy"
        ]
      ],
      "setup_steps": [
        "Connect one account first (e.g. Google Workspace, which carries the no-training commitment) and review the permission scopes granted.",
        "If training use matters to you, exercise the opt-out early - it applies go-forward, not retroactively.",
        "Run a low-stakes task (a lookup or draft) before enabling agent-to-agent trusted-network sharing."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Price not verified",
      "page_url": "https://agentcomparison.net/agents/instinct/"
    },
    {
      "id": "langgraph",
      "name": "LangGraph",
      "vendor": "LangChain",
      "kind": "framework",
      "category": "agent orchestration framework",
      "group": "builders",
      "summary": "LangChain documents LangGraph as a low-level runtime for long-running stateful agents, including durable execution and human-in-the-loop support.",
      "one_liner": "A developer framework for long-running, stateful agents.",
      "best_fit": "Developers implementing stateful agent workflows.",
      "poor_fit": "A finished assistant that works without development.",
      "decision_note": "A framework is a building block, not a ready-made assistant; compare developer requirements and runtime choices.",
      "availability": "Official documentation available; signup and plan/region details vary. See pricing/access notes.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Durable execution and persistence",
          "source_url": "https://docs.langchain.com/oss/python/langgraph/overview",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Human-in-the-loop state inspection",
          "source_url": "https://docs.langchain.com/oss/python/langgraph/overview",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Short- and long-term memory support",
          "source_url": "https://docs.langchain.com/oss/python/langgraph/overview",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Developers place dynamic interrupts before side effects and implement approval/rejection handling. Framework support alone does not enforce a universal permission policy."
      ],
      "limitations": [
        "A durable checkpointer and stable thread ID are prerequisites for resumable production workflows.",
        "Resuming restarts the interrupted node, so pre-interrupt side effects can repeat.",
        "Model, database, logging and hosting costs and data terms are separate from the framework."
      ],
      "runtime": [
        "Developer-chosen runtime; hosted products are separate"
      ],
      "privacy_security": "Workflow state can be persisted by the chosen checkpointer. Retention, access, deletion, model training and logs depend on the developer-selected services and configuration; no framework-wide privacy guarantee or retention period is established.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-22",
        "source_url": "https://docs.langchain.com/oss/python/langgraph/overview",
        "note": "LangGraph is an open-source framework (MIT licence, per the GitHub repository, re-verified 22 October 2026). This is not a promise of free agent operation: model inference, hosting and optional LangSmith services have separate costs. No complete deployed-workload price is verified."
      },
      "sources": [
        "https://docs.langchain.com/oss/python/langgraph/overview",
        "https://docs.langchain.com/oss/python/langgraph/checkpointers",
        "https://docs.langchain.com/oss/python/langgraph/interrupts",
        "https://github.com/langchain-ai/langgraph"
      ],
      "changelog": [
        {
          "date": "2026-10-22",
          "change": "Re-verified vendor documentation claims (durable execution, human-in-the-loop interrupts, persistence, memory) against current LangGraph docs; MIT licence and active maintenance confirmed via the GitHub repository API. No content corrections needed; no hands-on testing."
        },
        {
          "date": "2026-10-14",
          "change": "Expanded product-specific runtime, approval, access and privacy evidence with narrow setup steps. Documentation research, not hands-on testing."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        }
      ],
      "coverage_rows": [
        [
          "Product type",
          "Developer framework",
          "LangGraph is a low-level runtime for stateful agent workflows, not a finished assistant with ready-made accounts or permissions. Developers choose models, tools and deployment.",
          "https://docs.langchain.com/oss/python/langgraph/overview"
        ],
        [
          "Checkpoint storage",
          "Configure persistence",
          "Compile the graph with a checkpointer. In-memory examples are not durable production storage; SQLite, Postgres and other adapters have separate installation and operating needs.",
          "https://docs.langchain.com/oss/python/langgraph/checkpointers"
        ],
        [
          "Pause for input",
          "Implementation prerequisites",
          "Dynamic interrupt() needs a checkpointer, a thread ID and a JSON-serializable payload. It saves state and waits until the application resumes it; the developer decides where to pause.",
          "https://docs.langchain.com/oss/python/langgraph/interrupts"
        ],
        [
          "Resume identity",
          "Keep the same thread",
          "Command(resume=...) supplies the response to the interrupt. Reuse the thread ID to load the checkpoint; a new ID starts an empty thread rather than continuing the old one.",
          "https://docs.langchain.com/oss/python/langgraph/interrupts"
        ],
        [
          "Replay and side effects",
          "Node code can run again",
          "Resuming restarts the interrupted node from its beginning. Operations before interrupt can repeat, so use idempotent operations or separate side-effect nodes and deduplication.",
          "https://docs.langchain.com/oss/python/langgraph/interrupts"
        ],
        [
          "Human approval",
          "Not automatic for every tool",
          "Developers must put the interrupt before the actual action and handle approve, edit and reject responses. Static breakpoints are for debugging, not the recommended human-in-the-loop pattern.",
          "https://docs.langchain.com/oss/python/langgraph/interrupts"
        ],
        [
          "Licence and cost",
          "MIT is not free operation",
          "The repository lists an MIT licence. Model inference, hosting, storage and optional LangSmith services remain separate; no complete workload price is verified.",
          "https://github.com/langchain-ai/langgraph"
        ],
        [
          "Privacy and retention",
          "Your deployment decides",
          "Checkpointing persists workflow state; it is not a promise of deletion or model-training opt-out. Review the chosen store, model provider, logging and hosting terms. No framework-wide retention period is established.",
          "https://docs.langchain.com/oss/python/langgraph/checkpointers"
        ]
      ],
      "setup_steps": [
        "Build a small graph with a single read-only tool and explicit success/failure states.",
        "Choose a checkpointer and stable thread ID. Use a durable store for production rather than an in-memory demo.",
        "Place interrupt() before a side effect; implement approve, edit and reject responses with Command(resume=...).",
        "Check replay behavior and deduplicate writes before connecting real accounts. Review model, hosting and storage costs and data terms."
      ],
      "price_tags": [
        "free",
        "usage"
      ],
      "price_summary": "Open-source framework; model, hosting and LangSmith costs are separate",
      "page_url": "https://agentcomparison.net/agents/langgraph/"
    },
    {
      "id": "lucas",
      "name": "Lucas",
      "vendor": "Lucas Intelligence, Inc.",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Lucas describes a text-native iMessage assistant for email, calendar, documents, reminders and automations.",
      "one_liner": "An iMessage assistant for email, calendar, documents and reminders.",
      "best_fit": "Personal administration through iMessage.",
      "poor_fit": "Assuming a marketing summary proves every action asks for approval.",
      "decision_note": "Check proactive frequency and approval boundaries in the current service before connecting personal accounts.",
      "availability": "Official documentation available; signup and plan/region details vary. See pricing/access notes.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "iMessage interface",
          "source_url": "https://meetlucas.ai/faq",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Email, calendar, documents and reminders",
          "source_url": "https://meetlucas.ai/faq",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Requested briefings, nudges and follow-ups",
          "source_url": "https://meetlucas.ai/faq",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Connects via Google API Services: Gmail, Calendar, Contacts, Drive, Docs, Sheets, Slides, Tasks and account sign-in; the policy says new permissions require your authorization before access.",
        "With a grant, Lucas can access existing and historical data available through those APIs.",
        "Messaging channels: iMessage, WhatsApp and SMS, plus the Lucas web app.",
        "Revocation stops future access under that grant but does not itself delete information already received; deletion runs on account deletion or a request to privacy@meetlucas.ai."
      ],
      "limitations": [
        "Usage limits are not published as numbers",
        "Reliability, response speed and safe-action rates not tested"
      ],
      "runtime": [
        "Vendor-managed messaging assistant"
      ],
      "privacy_security": "Lucas's privacy policy states Google API data is 'never used to train generalized AI or machine learning models, regardless of other settings', and separately: 'We do not use your messages or other User Content to train AI or machine learning models' (de-identified/aggregated non-Google data may be used). Users can email privacy@meetlucas.ai with 'AI Training Opt-Out' for additional restrictions. Hidden/retained records persist until account deletion or a deletion request; disconnecting a service is separate from deletion. Checked 22 October 2026.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-22",
        "source_url": "https://meetlucas.ai/faq",
        "note": "Lucas's FAQ (read on 2 October 2026 after expanding the answers) says there is a free tier, Lucas Pro at $20/month or $192/year, and Lucas Max at $100/month or $960/year, with a seven-day reverse trial that steps down to Free unless you subscribe. It says usage limits apply from day one, gives no numeric allowances, and names the account billing page as the source of truth.",
        "class": "free_tier"
      },
      "sources": [
        "https://meetlucas.ai/faq",
        "https://meetlucas.ai/privacy-policy/"
      ],
      "changelog": [
        {
          "date": "2026-10-22",
          "change": "Field-level audit: FAQ pricing re-verified via browser (free, Pro $20/mo or $192/yr, Max $100/mo or $960/yr, 7-day reverse trial); privacy policy captured - no training on messages/User Content, Google API data never used for generalized AI training, email AI-training opt-out; Google API scopes (Gmail/Calendar/Contacts/Drive/Docs/Sheets/Slides/Tasks) with authorization-gated access; retention until deletion request. Coverage rows, permissions, setup steps, privacy fields filled. No hands-on testing."
        },
        {
          "date": "2026-10-22",
          "change": "Re-verified live FAQ pricing: free tier, Lucas Pro $20/month or $192/year, Lucas Max $100/month or $960/year, seven-day reverse trial stepping down to Free. Unchanged. No hands-on testing."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Free / Pro $20 / Max $100",
          "FAQ re-verified via browser (JS accordions): free tier; Pro $20/mo or $192/yr; Max $100/mo or $960/yr; 7-day reverse trial.",
          "https://meetlucas.ai/faq"
        ],
        [
          "Training use",
          "No training on user content",
          "Policy: no training on messages or User Content; Google API data never used for generalized AI training; email opt-out available for additional restrictions.",
          "https://meetlucas.ai/privacy-policy/"
        ],
        [
          "Channels",
          "iMessage, WhatsApp, SMS, web",
          "Two-way messaging across iMessage, WhatsApp and SMS plus a web app, per the privacy policy's service description.",
          "https://meetlucas.ai/privacy-policy/"
        ],
        [
          "Data retention",
          "Until deletion request",
          "Hidden records retained until account deletion or earlier request; disconnecting a service does not delete data already received.",
          "https://meetlucas.ai/privacy-policy/"
        ]
      ],
      "setup_steps": [
        "Connect one Google account and review each requested scope before granting - Lucas can read existing and historical data once granted.",
        "Send one low-risk scheduling or drafting request over SMS or iMessage and check the result before enabling automations.",
        "Note the deletion path: disconnecting a service does not delete stored data; deletion needs an account deletion or an email to privacy@meetlucas.ai."
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Free tier; Pro $20/month or $192/year; Max $100/month or $960/year; 7-day reverse trial",
      "page_url": "https://agentcomparison.net/agents/lucas/"
    },
    {
      "id": "manus",
      "name": "Manus",
      "vendor": "Manus",
      "kind": "product",
      "category": "general-purpose task execution",
      "group": "work-business",
      "summary": "Manus describes agent tools with a browser and file system that can execute tasks and create deliverables, including presentations and websites.",
      "one_liner": "A general task agent with a browser that produces files, slides and websites.",
      "best_fit": "Delivering task outputs rather than only answering questions.",
      "poor_fit": "Work requiring demonstrated reliability on sensitive actions.",
      "decision_note": "Evaluate finished output and intervention requirements, not the number of tools listed.",
      "availability": "Official documentation available; signup and plan/region details vary. See pricing/access notes.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Browser and file-system tools",
          "source_url": "https://www.manus.im/tools",
          "observed_at": "2026-10-01",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Presentations and websites",
          "source_url": "https://www.manus.im/tools",
          "observed_at": "2026-10-01",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Manus 2.0: Cascade engine, Manus Studio desktop workspace, Cloud Computer, Automations, Remote Control, Video Editor, Game Dev",
          "source_url": "https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0",
          "observed_at": "2026-10-01",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Cue: standalone app (invite code required) with a personal team of agents, each with its own email, phone number, wallet and computer",
          "source_url": "https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "No granular task-approval regime was found in the terms or help material reviewed; the acceptable-use rules prohibit sensitive personal data (SSNs, financial credentials, health data, precise location, others' personal data without consent) in prompts.",
        "Manus 2.0's Remote Control and Manus Studio work on your computer only in 'a connected and authorized desktop session' with apps and files you approve - an opt-in device session rather than a standing permission.",
        "Connecting third-party accounts (email, files) is your choice per feature; the terms put responsibility for having rights to share that account data on you."
      ],
      "limitations": [
        "General output execution documented; independent quality unknown"
      ],
      "runtime": [
        "Tasks run in the vendor environment"
      ],
      "privacy_security": "Privacy policy (last updated 21 April 2025) and Terms of Use (updated 28 November 2025), both read 22 October 2026. You own your Input and Output - Manus explicitly does not claim ownership - but you grant a perpetual, irrevocable license to use your content 'in an aggregated manner to improve the Services' and create usage data. Content is shared with third-party AI providers, and the terms warn those providers may retain rights to use or disclose your content and may not be required to keep it confidential. Retention is 'as long as necessary'; deletion can be requested and is processed on account termination per the retention policy. Analytics run through Amplitude (event data only per the policy). Vendor policy, not independently audited.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-22",
        "source_url": "https://help.manus.im/en/articles/11711111-what-is-the-current-membership-pricing-for-manus",
        "note": "Manus help article (read 22 October 2026) lists Free at $0/month (300 daily-refresh credits, Chat Mode and Manus 2.0 Lite), Pro from $20/month, a second Pro tier from $40/month with a 7-day free trial, and Team from $20/seat/month, with a 17% discount for annual billing. Confirm the checkout tier before subscribing."
      },
      "sources": [
        "https://www.manus.im/tools",
        "https://help.manus.im/en/articles/11711111-what-is-the-current-membership-pricing-for-manus",
        "https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0",
        "https://manus.im/privacy",
        "https://manus.im/terms"
      ],
      "changelog": [
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-09-29",
          "change": "Manus 2.0 documented in the Manus help center and reported by InfoWorld on 29 September 2026: Cascade engine, Manus Studio, Cloud Computer, Automations, Remote Control, Video Editor, Game Dev and the Cue app. Manus says its token, time and cost savings figures (23.2%, 28.2%, 32%) come from one tested configuration, not an average."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: privacy policy and terms read for the first time - user owns Input/Output but grants perpetual irrevocable licence for aggregated service improvement; third-party AI providers may retain rights over content; deletion on request. Cue standalone app capability added. Permissions field rewritten from unverified placeholder."
        }
      ],
      "price_tags": [
        "free",
        "listed",
        "usage"
      ],
      "price_summary": "Free with daily credits; paid plans from $20/month, Team from $20/seat/month (vendor help)",
      "page_url": "https://agentcomparison.net/agents/manus/"
    },
    {
      "id": "copilot-studio",
      "name": "Microsoft Copilot Studio",
      "vendor": "Microsoft",
      "kind": "platform",
      "category": "agent-building platform",
      "group": "builders",
      "summary": "Microsoft describes Copilot Studio as a no-code platform for creating and managing custom agents, workflows and apps on business data, published across channels.",
      "one_liner": "Microsoft's platform for building and governing agents on business data.",
      "best_fit": "Teams building business-data agents for several channels.",
      "poor_fit": "Comparing it as a ready-made personal assistant.",
      "decision_note": "Compare governance, connectors and operating cost with other builders, not consumer assistants.",
      "availability": "Standalone and Microsoft 365 Copilot-included rights differ by harness: Microsoft 365 Copilot includes the Copilot Chat and Standard harnesses for internal agents at no extra cost, while the standalone license adds the GitHub Copilot harness and external-channel publishing. Azure subscription, environment, connector policies and chosen channel determine available capabilities.",
      "last_checked": "2026-10-30",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Business-data connections",
          "source_url": "https://www.microsoft.com/en-us/microsoft-365-copilot/microsoft-copilot-studio",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Natural-language agent building",
          "source_url": "https://www.microsoft.com/en-us/microsoft-365-copilot/microsoft-copilot-studio",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Multi-channel publishing",
          "source_url": "https://www.microsoft.com/en-us/microsoft-365-copilot/microsoft-copilot-studio",
          "observed_at": "2026-10-12",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "User authentication and channel support must be selected; No authentication permits anyone with the link.",
        "Event triggers use maker credentials; autonomous actions require working maker authentication, which can expose data/actions to published users.",
        "Makers can require confirmation for tools that modify data. Admin data policies can block connectors, knowledge, HTTP, channels and event triggers."
      ],
      "limitations": [
        "Trigger payloads can consume metered usage; unattended does not mean free.",
        "Maker-authenticated triggers can expose maker-account data/actions.",
        "Publishing and auth changes must be tested in the intended environment and channel.",
        "Product-specific retention/training controls were not established in this feature check."
      ],
      "runtime": [
        "Managed business agent platform"
      ],
      "privacy_security": null,
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-30",
        "source_url": "https://www.microsoft.com/en-us/copilot/pricing/copilot-studio",
        "note": "Microsoft lists a tenant-wide license at USD $200/pack/month for 25,000 Copilot Credits; also a pay-as-you-go meter (requires an active Azure subscription linked to the environment) and a pre-purchase plan of Copilot Credit Commit Units with up to 20% savings and automatic pay-as-you-go backstop when prepaid credits run out. Actions/responses consume varying credits by usage. Microsoft 365 Copilot ($30/user) includes different, separate rights."
      },
      "sources": [
        "https://www.microsoft.com/en-us/copilot/products/copilot-studio",
        "https://www.microsoft.com/en-us/copilot/pricing/copilot-studio",
        "https://learn.microsoft.com/en-us/microsoft-copilot-studio/faqs-generative-orchestration",
        "https://learn.microsoft.com/en-us/microsoft-copilot-studio/authoring-triggers-about",
        "https://learn.microsoft.com/en-us/microsoft-copilot-studio/admin-data-loss-prevention",
        "https://learn.microsoft.com/en-us/microsoft-copilot-studio/configuration-end-user-authentication",
        "https://learn.microsoft.com/en-us/microsoft-copilot-studio/billing-licensing"
      ],
      "changelog": [
        {
          "date": "2026-10-12",
          "change": "Added maker, licensed internal-user and external-publication licence boundaries with a direct pricing-guide jump. Rechecked the listed primary sources; no universal free-runtime claim."
        },
        {
          "date": "2026-10-11",
          "change": "Expanded product-specific feature evidence and narrow setup from fresh official docs. Costs, credential exposure and limits remain explicit. Documentation research, not tested."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-10-26",
          "change": "CTR polish (non-test, near-page-1 work): meta title/description now answer the price query with the verified $200/pack/month figure; pricing fact re-verified on microsoft.com 26 Oct. checked_label unchanged - no full-record re-check."
        },
        {
          "date": "2026-10-30",
          "change": "Full-record re-check on microsoft.com pricing page + FAQ and product page: $200/pack/month for 25,000 Copilot Credits, pay-as-you-go via Azure and pre-purchase plan (up to 20% savings) all confirmed unchanged. New vendor framing recorded: rights organised by harness (Copilot Chat/Standard included with Microsoft 365 Copilot; GitHub Copilot harness and external channels standalone), free trial listed, Agent Builder and Copilot Studio Lite named. checked_label bumped (full-record check)."
        }
      ],
      "coverage_rows": [
        [
          "Building blocks",
          "Documented",
          "Generative orchestration selects configured topics, tools, agents and knowledge. It does not establish successful task completion.",
          "https://learn.microsoft.com/en-us/microsoft-copilot-studio/faqs-generative-orchestration"
        ],
        [
          "Event triggers",
          "Requires generative orchestration",
          "Outside events and recurrence can start work without a user prompt. Trigger availability depends on environment settings and data policies.",
          "https://learn.microsoft.com/en-us/microsoft-copilot-studio/authoring-triggers-about"
        ],
        [
          "Trigger credentials",
          "Maker account risk",
          "Triggers use maker credentials. Autonomous authenticated actions also need working maker authentication; published users may invoke data/actions using that access.",
          "https://learn.microsoft.com/en-us/microsoft-copilot-studio/authoring-triggers-about"
        ],
        [
          "Human confirmation",
          "Configurable",
          "Makers can require confirmation for tools that modify data. Do not assume all changes prompt merely because the agent is published.",
          "https://learn.microsoft.com/en-us/microsoft-copilot-studio/faqs-generative-orchestration"
        ],
        [
          "Data policies",
          "Admin configured",
          "Policies can block knowledge sources, connectors, HTTP, skills, channels and triggers. Data cannot be shared across connectors in different policy groups.",
          "https://learn.microsoft.com/en-us/microsoft-copilot-studio/admin-data-loss-prevention"
        ],
        [
          "User authentication",
          "Channel dependent",
          "No authentication lets anyone with the link interact. Microsoft authentication supports Teams/Microsoft 365; other authenticated channels need manual setup. Changes take effect after publication.",
          "https://learn.microsoft.com/en-us/microsoft-copilot-studio/configuration-end-user-authentication"
        ],
        [
          "Usage and external channels",
          "Licensed and metered",
          "The pricing page distinguishes included internal Microsoft 365 Copilot rights from standalone external channels. Packs, pay-as-you-go and prepurchase exist; Azure and overage coverage need review.",
          "https://www.microsoft.com/en-us/copilot/pricing/copilot-studio"
        ],
        [
          "Background billing",
          "Not free because unattended",
          "Trigger payloads count toward consumption when they trigger actions or responses, including recurring payloads.",
          "https://learn.microsoft.com/en-us/microsoft-copilot-studio/authoring-triggers-about"
        ]
      ],
      "setup_steps": [
        "Use a test environment and one read-only knowledge task.",
        "Check channel authentication, connector data-policy groups and whose credentials each action uses.",
        "Preview approved outputs before publication; require confirmation for risky tools.",
        "Add event triggers only after reviewing maker-account exposure, scope and usage budget."
      ],
      "price_tags": [
        "listed",
        "usage"
      ],
      "price_summary": "$200 per pack per month for 25,000 credits; pay-as-you-go and pre-purchase options",
      "licensing_note": "Microsoft's pricing FAQ (30 Oct 2026) frames rights by harness: Microsoft 365 Copilot includes Copilot Chat and Standard harness use for internal agents at no additional cost; the standalone license adds the GitHub Copilot harness, external-channel publishing and use by non-licensed users. Learn says the free maker-licence route needs tenant prepaid capacity; trial agents can be tested but not published. Microsoft 365 Copilot zero-rates specified internal classic/generative/tenant-grounded use for licensed users in Copilot Chat, Teams or SharePoint, not every action or every user. Microsoft lists a free trial and a lightweight in-context 'Agent Builder' inside Microsoft Copilot; a footnote limits some offers for agents built in Copilot Studio Lite.",
      "pricing_guide": {
        "href": "/guides/copilot-studio-pricing-and-licensing-explained/",
        "label": "Full Copilot Studio pricing and licensing guide"
      },
      "meta_title": "Copilot Studio pricing (Oct 2026): $200 pack + PAYG",
      "page_url": "https://agentcomparison.net/agents/copilot-studio/"
    },
    {
      "id": "muse",
      "name": "Muse",
      "vendor": "Meta",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Meta documents Muse as a personal agent for actions, reminders, ongoing conversations and file work.",
      "one_liner": "Meta's personal agent for reminders, actions and file work.",
      "best_fit": "Personal reminders, ongoing tasks and artifact work.",
      "poor_fit": "Unmonitored actions where errors would be costly.",
      "decision_note": "Review actions and define boundaries: Meta warns that Muse may be inaccurate or take unexpected actions. Pricing is now documented (free weekly limit; Power $20/mo; Maximum $100/mo) but regional availability is not - check the app shows Muse as available before recommending. Muse is not in our text-message assistants table under the published inclusion rule: Meta documents conversations in its own app and web chat only, with no messaging-service channel (iMessage, SMS, WhatsApp, Telegram, Slack, email or phone).",
      "availability": "Muse app (iOS/Android) and muse.ai under a Meta Account (Facebook/Instagram/email/phone signup). Meta says Muse and its subscriptions are in limited testing and are not available in all locations; no supported-country list is published on its help pages (Japan rollout 23 Oct and Singapore 28 Oct 2026 reported by press, not yet listed on vendor pages). Signup eligibility by country was not tested.",
      "last_checked": "2026-10-28",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Reminders and ongoing conversation context",
          "source_url": "https://www.meta.com/help/artificial-intelligence/1331373868832401/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "PDFs, spreadsheets and other artifacts",
          "source_url": "https://www.meta.com/help/artificial-intelligence/1331373868832401/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "App and web access",
          "source_url": "https://www.meta.com/help/artificial-intelligence/1331373868832401/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Connectors: Meta and third-party app connections (Gmail/calendar-style read and write), retrieve-only mode per connector, custom connectors via chat, credentials in Secure Credentials Store. Facebook, Instagram and Threads auto-connect via Accounts Center.",
          "source_url": "https://www.meta.com/help/superintelligence/1687253048996149/",
          "observed_at": "2026-10-28",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": null,
          "region_restrictions": "Limited testing; not all locations"
        },
        {
          "name": "Built-in web browser: booking appointments, purchases, online forms and price-watching, with a live viewing window, take-control pause and stop. Browsing credentials stored in the Secure Credentials Store.",
          "source_url": "https://www.meta.com/help/artificial-intelligence/2124746764949121/",
          "observed_at": "2026-10-28",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": null,
          "region_restrictions": "Limited testing; not all locations"
        },
        {
          "name": "Approval system: per-connector read/write gates, allow once/task/site/always/deny decisions, activity log, scheduled-task approvals and allowed-website list.",
          "source_url": "https://www.meta.com/help/artificial-intelligence/1385290430137537/",
          "observed_at": "2026-10-28",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Memory and identity files: Memories plus editable Memory.md, Soul.md and Identity.md; a forget skill erases topics; full reset permanently deletes chats, files and active tasks. Goals tab tracks personal goals with check-ins.",
          "source_url": "https://www.meta.com/help/artificial-intelligence/2225571704857152/",
          "observed_at": "2026-10-28",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Runs from the Muse app or web under a Meta Account (Facebook/Instagram/email/phone signup paths).",
        "Connectors for Meta and third-party apps (email and calendar examples; Facebook, Instagram and Threads auto-connect via Accounts Center). Per-connector read vs write can be set to retrieve-only. Custom connectors can be created by asking Muse; Meta does not review them.",
        "Granular approvals: per connector 'Ask for some actions' (write and important reads) or 'Always ask'; web-access defaults; approval options are Allow once / for this task / for this site / Always allow / Deny. Meta says Muse will not take many important actions, like sending an email, without approval by default.",
        "Built-in browser for bookings, purchases, form-filling and price-watching; user can watch live, take control (pause) or stop the task. Credentials sit in a Secure Credentials Store the model does not see.",
        "Activity log, scheduled-task and artifact approvals, and an allowed-websites list are user-manageable in Settings.",
        "Can send emails, texts and artifacts to others; Meta suggests asking it to attribute AI-generated content to the agent. Some actions (like a sent email) cannot be reversed.",
        "Creates and edits artifacts (PDFs, spreadsheets, documents, images, podcasts) saved to a Library workspace."
      ],
      "limitations": [
        "Meta warns of inaccurate responses or unexpected actions and tells users to review activity regularly.",
        "Some actions, like a sent email, cannot be reversed.",
        "AI-training opt-out is a manual setting that starts ON.",
        "Custom connectors are not reviewed by Meta.",
        "Limited testing: no supported-country list published; subscription management is locked to the original platform, device type and store account.",
        "Free usage limit size and token-to-task equivalence are unpublished. Reliability and task success were not tested."
      ],
      "runtime": [
        "Meta account; iOS/Android app or muse.ai web",
        "Per-user cloud VM with a built-in browser the user and Muse can both control",
        "Side chats share context with the main chat"
      ],
      "privacy_security": "Meta's Muse help pages (re-read 28 October 2026) state: each user's Muse runs in an isolated cloud VM, with a Confidential VM option planned that Meta says is designed to keep information private even from Meta. Muse does not share conversations or VM data with Meta ad systems. Credentials live in a Secure Credentials Store and are used only after checks that the service, authorization and permission type match; the AI model does not see passwords. Meta documents prompt-injection mitigations (model training, automated detection, permission checks operating separately from the model). AI-training on interactions is ON by default (Settings > Data controls); turning it off stops use for model improvement and applies to previous interactions; when on, Meta says it removes certain PII categories and disassociates interactions from the account. Connector data exchange is scoped to what the task needs (email search does not download the whole inbox unless told); Google Workspace API use adheres to the Google User Data Policy Limited Use requirements. Custom connectors are not reviewed by Meta - grant access with caution.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": "Free: weekly usage limit (size unpublished). Power: 500M Muse tokens/week. Maximum: 3B Muse tokens/week.",
        "usage_charges": null,
        "verified_at": "2026-10-28",
        "source_url": "https://www.meta.com/help/subscriptions/1021145227643680/",
        "note": "Meta's subscription help pages (read 28 October 2026): free with a weekly usage limit that refreshes; Power $20/month (500M Muse tokens/week); Maximum $100/month (3B tokens/week). Subscribe at muse.ai or in the mobile app; manage and cancel on the same platform, device type and store account used to subscribe. Meta says Muse and subscriptions are in limited testing and not available in all locations.",
        "class": "free"
      },
      "sources": [
        "https://www.meta.com/help/artificial-intelligence/1331373868832401/",
        "https://www.meta.com/help/subscriptions/1021145227643680/",
        "https://www.meta.com/help/superintelligence/1687253048996149/",
        "https://www.meta.com/help/artificial-intelligence/2124746764949121/",
        "https://www.meta.com/help/artificial-intelligence/1385290430137537/",
        "https://www.meta.com/help/superintelligence/1047255454427887/",
        "https://www.meta.com/help/artificial-intelligence/2225571704857152/",
        "https://www.meta.com/help/subscriptions/28510110068583598/"
      ],
      "changelog": [
        {
          "date": "2026-10-28",
          "change": "Full record re-check (9 Meta help pages read: getting-started, subscriptions x2, connectors, web browsing, guidance/approval, privacy/security, manage data, help index). Pricing established: free weekly limit; Power $20/mo (500M tokens/week); Maximum $100/mo (3B tokens/week); class free->free with listed tags. Added connectors, built-in browser, approval system, memory/identity files capabilities; rewrote privacy_security (isolated VM, no ads sharing, Secure Credentials Store, prompt-injection mitigations, training toggle default-on). Availability: limited testing, no country list published; Japan 23 Oct and Singapore 28 Oct rollouts noted as press-reported, not vendor-listed. checked_label bumped to 28 October 2026. No hands-on testing."
        },
        {
          "date": "2026-10-23",
          "change": "Field-level audit from Meta help documentation: free with usage limit (paid upgrade, price unpublished); Meta Account sign-in; app connections and directed actions; official safety posture - may be inaccurate or take unexpected actions, monitor and intervene, boundaries via user instructions; artifact creation (PDFs, spreadsheets, images, podcasts) in Library; context memory on by design; AI-attribution suggestion for outbound content. Coverage rows, permissions, setup steps, privacy fields filled. No hands-on testing."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-09-29",
          "change": "Meta announced Muse for Small Business, adding business skills and connectors. Meta says nothing publishes, sends or spends without approval. Source: https://about.fb.com/news/2026/09/introducing-muse-small-business/"
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Free with usage limit",
          "Meta help: Muse is free with a usage limit; hitting the limit means upgrading to a paid subscription or waiting for the refresh. No paid-tier price published on the help pages checked.",
          "https://www.meta.com/help/artificial-intelligence/1331373868832401/"
        ],
        [
          "Account model",
          "Meta Account",
          "App or web access via Meta Account; alternative signup via Facebook, Instagram, email or phone number.",
          "https://www.meta.com/help/artificial-intelligence/1331373868832401/"
        ],
        [
          "Safety posture",
          "User-monitored",
          "Meta's own documentation: Muse 'may be inaccurate or take unexpected actions - monitor its actions carefully and intervene as needed.' Boundary-setting is left to user instructions.",
          "https://www.meta.com/help/artificial-intelligence/1331373868832401/"
        ],
        [
          "Memory",
          "On by design",
          "Remembers conversation context across main chat and side chats; name, personality and preferences are customizable by instruction.",
          "https://www.meta.com/help/artificial-intelligence/1331373868832401/"
        ]
      ],
      "setup_steps": [
        "Sign in with a Meta Account (or create one via Facebook/Instagram/email/phone) and start on the Chat tab.",
        "State boundaries explicitly in your first instructions (what it may search, send or change) - Meta's own guidance relies on user-set limits.",
        "Review its first task outputs and enable the AI-attribution note before letting it email or text anyone on your behalf."
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Free with weekly usage limit; Power $20/month; Maximum $100/month",
      "page_url": "https://agentcomparison.net/agents/muse/"
    },
    {
      "id": "openai-dots",
      "name": "OpenAI dots",
      "vendor": "OpenAI",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "OpenAI dots uses shared ChatGPT/plugin permissions and memory for connected work, proactive private research, and permitted local or cloud computer tasks. Its FAQ distinguishes read-only proactive research, action approvals, Custom Rules and takeover, while Enterprise workspace controls separately govern access and messaging.",
      "one_liner": "OpenAI's personal assistant that uses plugins and shared ChatGPT memory.",
      "best_fit": "Personal connected-app work with shared ChatGPT context.",
      "poor_fit": "A user needing per-memory inspection and deletion.",
      "decision_note": "Disconnecting a plugin stops new access but does not delete context already retained.",
      "availability": "OpenAI's getting-started article (vendor-updated 28 October 2026) documents dots rolling out gradually to ChatGPT Pro users aged 18+ in markets excluding the EEA, Switzerland and the UK, and to Business Premium users in supported regions subject to workspace eligibility. Enterprise, Edu and Healthcare can try the beta when a workspace admin enables it (off by default); access may take several days. Setup on iOS, Android, desktop app or desktop web; not supported on mobile web.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Shared connections: Plugin permissions apply across dots, ChatGPT, Work and Codex; enabling dots does not grant all apps. Manage actual plugin scopes separately.",
          "source_url": "https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs",
          "observed_at": "2026-10-09",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Account, workspace grants and scopes",
          "region_restrictions": "Not established"
        },
        {
          "name": "Proactive research: Permitted sources can feed private notes, but research tools cannot directly send messages, change plugin content or control browser/computer. Follow-up actions use normal authorization.",
          "source_url": "https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs",
          "observed_at": "2026-10-09",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Account, workspace grants and scopes",
          "region_restrictions": "Not established"
        },
        {
          "name": "Local device access: Connect the device and grant camera/mic/screen access in device settings. Enterprise local command/file access additionally needs admin enablement; app must be open and device online.",
          "source_url": "https://help.openai.com/en/articles/20001554-manage-dots-in-chatgpt-workspaces",
          "observed_at": "2026-10-09",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Account, workspace grants and scopes",
          "region_restrictions": "Not established"
        },
        {
          "name": "Cloud computer: Browser, network, desktop and password-manager capabilities are separately managed. Cloud does not inherit local VPN, sign-ins or device policies.",
          "source_url": "https://help.openai.com/en/articles/20001554-manage-dots-in-chatgpt-workspaces",
          "observed_at": "2026-10-09",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Account, workspace grants and scopes",
          "region_restrictions": "Not established"
        },
        {
          "name": "Rules and approvals: Custom Rules support boundaries but cannot remove core safeguards/Autoreview. Sensitive transfers/password changes require takeover; some messages and purchases accept specifically scoped advance approval.",
          "source_url": "https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs",
          "observed_at": "2026-10-09",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Account, workspace grants and scopes",
          "region_restrictions": "Not established"
        },
        {
          "name": "Messaging channels: the desktop app connects messaging channels such as Slack or texting; dots are not supported on mobile web.",
          "source_url": "https://help.openai.com/en/articles/20001530-getting-started-with-your-dot",
          "observed_at": "2026-10-29",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Pro / Business Premium rollout",
          "region_restrictions": "Excludes EEA, Switzerland, UK per vendor"
        }
      ],
      "permissions": [
        "Plugin permissions are shared across dots, ChatGPT, ChatGPT Work and Codex.",
        "Custom Rules add boundaries but cannot override core safeguards. Sensitive actions can require takeover or approval; scoped advance approval is supported for some actions.",
        "Enterprise access is off by default; app authorization and local/cloud permissions are separate.",
        "Only owner instructions direct an Enterprise dot; other people's Slack DMs/mentions do not start work. Posts in channels are visible to members.",
        "Proactive research tools cannot directly message others, write through plugins or control browser/computer. A follow-up action still needs normal authorization; Custom Rules cannot remove this restriction.",
        "Password changes and money transfers require takeover. Saved-card purchases and recurring messages may use narrowly scoped advance approval, not open-ended ongoing authority."
      ],
      "limitations": [
        "Individual dot memories cannot currently be viewed or edited",
        "Disconnecting services does not delete retained context",
        "Phone messaging is unavailable for Enterprise at launch, according to workspace docs."
      ],
      "runtime": [
        "ChatGPT-connected personal assistant with permitted plugins and shared context",
        "Optional local computer and separately enabled cloud browser/network/desktop capabilities",
        "Enterprise Slack/Teams setup where supported; phone messaging unavailable at Enterprise launch"
      ],
      "privacy_security": "OpenAI says content is encrypted in transit and at rest. Disconnecting stops new plugin access, not deletion of retained context. Delete the dot to delete its context; separately stored files, conversations and ChatGPT Memory remain separate. Individual dot memories cannot currently be viewed or edited. Business, Enterprise and Edu data is not used for training by default; personal plans use the Improve the model for everyone setting. Human safety review can still occur. Vendor statements, not an audit. Proactive notes are not directly used for training, but information brought into eligible conversations/tasks may be, depending on settings. Turning off shared Memory does not erase received context. Supported secure-login forms differ from passwords sent in chat/plugin content.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-29",
        "source_url": "https://help.openai.com/en/articles/20001530-getting-started-with-your-dot",
        "note": "OpenAI's getting-started article (vendor-updated 28 October 2026, read 29 October 2026) states the first dot is included in the Pro or Business Premium plan at no extra cost, with a plan allowance for deeper work and extended limits for the first month after launch; OpenAI says additional dots and output scaling will be offered in future. No standalone dots price and no Pro plan dollar figure appears in the article.",
        "class": "see_vendor"
      },
      "sources": [
        "https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs",
        "https://help.openai.com/en/articles/20001554-manage-dots-in-chatgpt-workspaces",
        "https://help.openai.com/en/articles/20001530-getting-started-with-your-dot"
      ],
      "changelog": [
        {
          "date": "2026-10-09",
          "change": "Re-read official features, permissions, pricing and privacy sources; added product-specific evidence table and narrow-start steps. Vendor documentation, no hands-on tests."
        },
        {
          "date": "2026-10-07",
          "change": "Rechecked public FAQ and workspace guidance; clarified training settings, deletion and approval scope."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-10-23",
          "note": "Fresh source-to-field re-check (phase 2): dots privacy/safety FAQ (vendor-updated 21 Oct) re-read in full - approval defaults, proactive-research limits, memory sharing, training controls (Business/Enterprise/Edu off by default; personal plans via 'Improve the model for everyone'), deletion model and human-review carve-out all still match the record. Counts toward the detailed ledger again."
        },
        {
          "date": "2026-10-29",
          "change": "Field-level audit: OpenAI getting-started help article (vendor-updated 28 Oct) read - rollout documented for Pro 18+ (excl. EEA/CH/UK) and Business Premium (supported regions), Enterprise beta admin-enabled; first dot included in plan at no extra cost with deeper-work allowance; messaging channels (Slack, texting) via desktop; availability/pricing/capabilities/meta updated. checked_label unchanged (not a full-record check)."
        }
      ],
      "coverage_rows": [
        [
          "Shared connections",
          "Documented",
          "Plugin permissions apply across dots, ChatGPT, Work and Codex; enabling dots does not grant all apps. Manage actual plugin scopes separately.",
          "https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs"
        ],
        [
          "Proactive research",
          "Read/notes only",
          "Permitted sources can feed private notes, but research tools cannot directly send messages, change plugin content or control browser/computer. Follow-up actions use normal authorization.",
          "https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs"
        ],
        [
          "Local device access",
          "Separate permission",
          "Connect the device and grant camera/mic/screen access in device settings. Enterprise local command/file access additionally needs admin enablement; app must be open and device online.",
          "https://help.openai.com/en/articles/20001554-manage-dots-in-chatgpt-workspaces"
        ],
        [
          "Cloud computer",
          "Distinct controls",
          "Browser, network, desktop and password-manager capabilities are separately managed. Cloud does not inherit local VPN, sign-ins or device policies.",
          "https://help.openai.com/en/articles/20001554-manage-dots-in-chatgpt-workspaces"
        ],
        [
          "Rules and approvals",
          "Not universal ask-first",
          "Custom Rules support boundaries but cannot remove core safeguards/Autoreview. Sensitive transfers/password changes require takeover; some messages and purchases accept specifically scoped advance approval.",
          "https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs"
        ],
        [
          "Memory",
          "Deletion distinction",
          "Shares ChatGPT memories/context; turning Memory off stops sharing, not existing context. Individual dot memories cannot be inspected/edited/deleted individually.",
          "https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs"
        ],
        [
          "Delete/disconnect",
          "Different effects",
          "Disconnect stops new access. Delete dot removes its own context; files, Codex threads, ChatGPT conversations and Memory remain separate.",
          "https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs"
        ],
        [
          "Training",
          "Plan/setting-dependent",
          "Personal plans use Improve the model for everyone; Business/Enterprise/Edu not trained by default. Proactive notes are not directly trained on, but their content in eligible work may be.",
          "https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs"
        ],
        [
          "Work messaging",
          "Enterprise caveat",
          "Supported Slack/Teams setup needs admin grants and member connection; Slack may require app approval. Only owner directs dot; others' mentions/DMs do not start work. Phone iMessage/RCS/WhatsApp unavailable at Enterprise launch.",
          "https://help.openai.com/en/articles/20001554-manage-dots-in-chatgpt-workspaces"
        ],
        [
          "Credentials and purchases",
          "Supported-flow limits",
          "Secure login form keeps supported credentials from the model; chat/plugin passwords are not covered. Auth checks may need takeover; saved-merchant-card purchases require scoped approval.",
          "https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs"
        ],
        [
          "Price and eligibility",
          "Unresolved",
          "Enterprise Beta is off by default; age 18+ per FAQ. Standalone pricing, complete personal-plan access and regions remain unconfirmed.",
          "https://help.openai.com/en/articles/20001554-manage-dots-in-chatgpt-workspaces"
        ]
      ],
      "setup_steps": [
        "Confirm your actual account has dots access; published Enterprise guidance is not proof of personal-plan eligibility or a free standalone service.",
        "Enterprise admins should review Use dots (Beta), applicable custom roles, local/cloud capabilities and app authorizations separately. Default Enterprise access is off.",
        "Review shared plugin access in ChatGPT and set narrow Custom Rules before delegating a task. Select a draft/read-only task first.",
        "If using a computer, connect it and grant only necessary device permissions; distinguish local access from cloud sessions and their separate sign-ins.",
        "Review Activity View and stop/change instructions as needed. Before removing access, separately consider dot context, stored files/conversations and ChatGPT Memory; disconnecting alone is not deletion."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "First dot included in ChatGPT Pro / Business Premium (no extra cost, per vendor); deeper-work allowance in plan",
      "meta_title": "OpenAI dots (Oct 2026): beta access, memory and controls",
      "page_url": "https://agentcomparison.net/agents/openai-dots/"
    },
    {
      "id": "poke",
      "name": "Poke",
      "vendor": "The Interaction Company of California",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Poke documents an assistant in Apple Messages, Telegram, WhatsApp and RCS for email, scheduling, reminders and web search.",
      "one_liner": "A messaging-app assistant for email, scheduling, reminders and web search.",
      "best_fit": "Messaging-based email, calendar and reminder work.",
      "poor_fit": "Granting broad access before checking approval controls.",
      "decision_note": "Poke documents messaging-based email, scheduling, reminders and connected-app work. Before connecting private accounts, review both integration scopes and Maximum Privacy: the general privacy notice permits model training unless that setting is selected. Numeric rate limits and universal write-approval behavior remain unverified.",
      "availability": "Documentation lists Apple Messages, Telegram, WhatsApp and RCS, with worldwide availability claimed by the vendor. Actual signup and task reliability were not tested.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Email reading, search and drafting",
          "source_url": "https://poke.com/docs",
          "observed_at": "2026-10-01",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Calendar scheduling and availability",
          "source_url": "https://poke.com/docs",
          "observed_at": "2026-10-01",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Reminders and web search",
          "source_url": "https://poke.com/docs",
          "observed_at": "2026-10-01",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Apple Messages, Telegram, WhatsApp and RCS",
          "source_url": "https://poke.com/docs",
          "observed_at": "2026-10-01",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "You explicitly connect integrations through a Recipe or an MCP-compatible server. Supported integrations can read information and perform actions; the pages reviewed do not establish a universal confirmation-before-every-write rule.",
        "Review the connected account and scopes, and disconnect it from the integrations page when no longer needed. Review Maximum Privacy separately from integration access."
      ],
      "limitations": [
        "Exact numeric rate limits, included credits and overage rates were not established from the pricing page.",
        "The monthly pricing view lists Free $0, Pro $19 and Ultra $199. Annual prices were not checked in this pass.",
        "Integration docs show read and write features, but do not establish a universal per-action approval gate.",
        "Maximum Privacy excludes model training under the policy; it does not establish zero retention or no provider processing.",
        "Reliability, response speed and safe-action rates were not tested."
      ],
      "runtime": [
        "Vendor-managed messaging assistant"
      ],
      "privacy_security": "The privacy notice, dated 11 September 2026, permits data use for model training but excludes users who select Maximum Privacy. Google API data has separate Limited Use, advertising and human-access restrictions. The notice describes purpose- and legal-based retention rather than a fixed number of days; data may be deleted, de-identified or isolated in backups pending deletion. Maximum Privacy is not a claim that no processing or retention occurs.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-23",
        "source_url": "https://poke.com/pricing",
        "note": "Poke's pricing page (read 22 October 2026, monthly view) lists Free at $0 with no credit card, Pro at $19/month and Ultra at $199/month, billed monthly. Pro adds frontier models, real-time background automations, higher rate limits and priority support. Ultra adds the highest rate limits and pay-as-you-go usage beyond included credits. No numeric limits are shown, and a yearly toggle exists that we did not read.",
        "class": "free"
      },
      "sources": [
        "https://poke.com/docs",
        "https://poke.com/docs/managing-integrations",
        "https://poke.com/pricing",
        "https://poke.com/privacy"
      ],
      "changelog": [
        {
          "date": "2026-10-16",
          "change": "Added eight evidence rows and narrow setup guidance; rechecked monthly pricing, integration scope and privacy policy, including Maximum Privacy training opt-out and unknown fixed retention limits. No hands-on tests."
        },
        {
          "date": "2026-10-06",
          "change": "Added Poke vs Instinct comparison; re-read Poke pricing, docs and privacy notice and Instinct terms. No change to the record above."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-10-23",
          "note": "Fresh source-to-field re-check (phase 2): pricing re-verified - Free $0 no card, Pro $19/mo (frontier models, background automations, higher rate limits), Ultra $199/mo (frontier on every action, pay-as-you-go beyond credits). Privacy notice re-read: training permitted except Maximum Privacy users; Google Limited Use restrictions - record statements still accurate. Counts toward the detailed ledger again."
        }
      ],
      "coverage_rows": [
        [
          "Messaging channels",
          "Documented",
          "Apple Messages, Telegram, WhatsApp and RCS. Vendor claims worldwide availability; signup was not tested.",
          "https://poke.com/docs"
        ],
        [
          "Email and calendar",
          "Documented",
          "Read, search and draft email; schedule meetings and check availability after connecting accounts.",
          "https://poke.com/docs"
        ],
        [
          "Reminders and web research",
          "Documented",
          "Natural-language reminders and web search are listed. Delivery timing and answer quality were not tested.",
          "https://poke.com/docs"
        ],
        [
          "Recipes and connected apps",
          "Read and write scope",
          "Recipes can connect apps such as Linear, Notion and GitHub. Check OAuth or API-key scope before enabling them.",
          "https://poke.com/docs/managing-integrations"
        ],
        [
          "Custom MCP and disconnect",
          "Documented",
          "Docs support MCP-compatible server connections and an integration disconnect flow. Server-specific safety is not verified.",
          "https://poke.com/docs/managing-integrations"
        ],
        [
          "Human approval",
          "Not established universally",
          "The reviewed docs do not promise a fresh approval before every external write. Do not infer this from natural-language chat.",
          "https://poke.com/docs/managing-integrations"
        ],
        [
          "Training and retention",
          "Policy with opt-out",
          "Maximum Privacy excludes model training. Retention depends on purpose and legal needs; no fixed all-data deletion period is stated.",
          "https://poke.com/privacy"
        ],
        [
          "Pricing and headroom",
          "Monthly view only",
          "Free $0, Pro $19/month, Ultra $199/month. Ultra mentions pay-as-you-go beyond credits, but exact limits and overage rates are not established.",
          "https://poke.com/pricing"
        ]
      ],
      "setup_steps": [
        "Choose a supported messaging channel and review the current plan and usage terms.",
        "Review Maximum Privacy and the privacy notice before connecting personal inboxes or files.",
        "Connect one low-risk integration through a Recipe. Check the actual account and requested scopes.",
        "Start with a read-only request that has an answer you can verify. Do not assume write approval is universal.",
        "Inspect results and usage before adding automations. Disconnect the integration when finished or if its scope is too broad."
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Free $0; Pro $19/month; Ultra $199/month (monthly billing view)",
      "page_url": "https://agentcomparison.net/agents/poke/"
    },
    {
      "id": "replit-agent",
      "name": "Replit Agent",
      "vendor": "Replit",
      "kind": "product",
      "category": "application building agent",
      "group": "coding-agents",
      "summary": "Replit documents an agent that creates applications, sets up projects, checks its work and fixes problems from natural-language instructions.",
      "one_liner": "Builds and fixes whole apps from plain-language instructions inside Replit.",
      "best_fit": "Building apps and related artifacts from natural-language requests.",
      "poor_fit": "Assuming a prototype is production-ready without review.",
      "decision_note": "Separate building an app from maintaining a production app; record deployment and usage charges before recommending a plan.",
      "availability": "Starter has daily-capped chat/Free Mode but no full build, Plan Mode, connectors or background tasks. Core and Pro support these with different allowances; Enterprise has separate Auto controls.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Writes code, sets up infrastructure and tests",
          "source_url": "https://docs.replit.com/features/agent/overview",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Plan mode before changes",
          "source_url": "https://docs.replit.com/features/agent/overview",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Checkpoints and rollback",
          "source_url": "https://docs.replit.com/features/agent/overview",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Privacy controls are workspace/app-level (public vs private apps); nothing more granular was found in the policy text.",
        "Agent and voice features run through third-party ML providers per the privacy policy.",
        "Do Not Sell opt-out link at replit.com footer; Google Analytics opt-out via Google's browser plug-in."
      ],
      "limitations": [
        "Generated work still needs code, deployment and security review.",
        "Free Mode is allowance-limited, not proof the subscription, deployment or API calls are free.",
        "Usage dashboard can lag up to 30 minutes; set billing caps before paid sessions.",
        "Product-specific data retention/training controls were not established by these two checked pages."
      ],
      "runtime": [
        "Replit project environment"
      ],
      "privacy_security": "Privacy policy (updated 21 October 2026) and Terms of Service read 22 October 2026. Default posture is social: your profile, code and forum posts are visible to other users and can be indexed by search engines BY DEFAULT; content in private apps is not made public, but Replit reserves the right to access private-app content for troubleshooting, improving the service, and safety/security. Service providers processing data include machine-learning providers powering the Agent and voice dictation. The privacy policy does not state explicitly whether user code trains AI models - no training commitment either way was found in the 21 October 2026 text. Advertising data-sharing may constitute a 'sale'/'sharing' under US state laws, with a Do Not Sell opt-out link. Terms include binding individual arbitration with a 30-day opt-out, and grant Replit broad rights to host and modify your content to operate the service. Vendor policy, not independently audited.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-22",
        "source_url": "https://docs.replit.com/billing/ai-billing",
        "note": "Free Mode supports no-cost work within a plan allowance; Core/Pro allowances reset every five hours and have weekly limits. Confirmed paid work in Power or Max uses effort-based pricing. Paid-path Plan Mode can bill without code changes. Credits also cover cloud services and third-party model APIs. No fixed per-request total or current subscription price is claimed."
      },
      "sources": [
        "https://docs.replit.com/features/agent/overview",
        "https://docs.replit.com/billing/ai-billing",
        "https://replit.com/privacy",
        "https://replit.com/terms"
      ],
      "changelog": [
        {
          "date": "2026-10-11",
          "change": "Expanded product-specific feature evidence and narrow setup from fresh official docs. Costs, credential exposure and limits remain explicit. Documentation research, not tested."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-09-25",
          "change": "Replit announced it acquired Atta and added interactive charts in Replit chat (Replit blog)."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: privacy policy (21 October 2026) and terms read - public-by-default posture, private-app access reservation, ML providers for Agent/voice, no explicit AI-training statement found, Do Not Sell and arbitration opt-outs documented. Permissions field filled from placeholder."
        }
      ],
      "coverage_rows": [
        [
          "Plan before changes",
          "Documented",
          "Plan Mode maps out work before code/data changes. Paid-path planning can bill even when no code changes.",
          "https://docs.replit.com/billing/ai-billing"
        ],
        [
          "Free versus paid modes",
          "Allowance and approval matter",
          "Free Mode supports everyday work within an allowance. Core/Pro reset every five hours with weekly limits; paid Power/Max work requires confirmation. Enterprise uses a separate Auto setup.",
          "https://docs.replit.com/features/agent/overview"
        ],
        [
          "Starter capabilities",
          "Limited",
          "Current billing table lists daily-capped chat/Free Mode, but no full build, Plan Mode, connectors or background tasks on Starter.",
          "https://docs.replit.com/billing/ai-billing"
        ],
        [
          "Artifacts and apps",
          "Documented",
          "Web/mobile apps, slides, designs and data outputs can share a project/backend. Agent writes infrastructure and tests; this is not proof a generated app is production-ready.",
          "https://docs.replit.com/features/agent/overview"
        ],
        [
          "Checkpoints",
          "Reviewable rollback",
          "Completed requests create checkpoints. Vendor says these support rollback; verify what is restored rather than assuming an external payment or message is undone.",
          "https://docs.replit.com/features/agent/overview"
        ],
        [
          "Background capacity",
          "Plan-dependent",
          "Billing docs list 0 active tasks on Starter, 1 on Core and 10 on Pro.",
          "https://docs.replit.com/billing/ai-billing"
        ],
        [
          "Paid effort and APIs",
          "Variable cost",
          "Paid checkpoints reflect request complexity. Third-party model services can deduct public API rates from credits; no fixed per-request total is claimed.",
          "https://docs.replit.com/billing/ai-billing"
        ],
        [
          "Budget visibility",
          "Set caps before paid work",
          "Credits also cover cloud services. Alerts and hard budget caps are documented; usage dashboard data may lag up to 30 minutes.",
          "https://docs.replit.com/billing/ai-billing"
        ]
      ],
      "setup_steps": [
        "Start with Free Mode and check the actual allowance in Settings > Usage.",
        "Describe one small task and review a plan before allowing code/data changes.",
        "Review the diff, test output and checkpoint after completion; do not assume rollback reverses external actions.",
        "Before Power/Max or deployment, inspect paid-action confirmation, cloud/model costs and billing caps."
      ],
      "price_tags": [
        "usage"
      ],
      "price_summary": "Free Mode within allowance; paid Power/Max work and cloud usage vary",
      "page_url": "https://agentcomparison.net/agents/replit-agent/"
    },
    {
      "id": "agentforce",
      "name": "Salesforce Agentforce",
      "vendor": "Salesforce",
      "kind": "platform",
      "category": "enterprise agent platform",
      "group": "work-business",
      "summary": "Salesforce describes Agentforce as a platform for autonomous enterprise support across web, phone and applications.",
      "one_liner": "Salesforce's platform for autonomous support agents across web, phone and apps.",
      "best_fit": "Enterprise agents connected to Salesforce workflows.",
      "poor_fit": "Treating enterprise deployment as a consumer assistant.",
      "decision_note": "Identify the customer-facing or employee use case, licence prerequisites and org consumption model before budgeting. An employee access licence is not necessarily unlimited agent usage; a standard action is not a conversation.",
      "availability": "Choose the actual customer-facing or employee product and supported Salesforce edition first. The Service Agent readiness guide lists Enterprise, Performance, Unlimited and Developer Editions with Foundations or Agentforce 1 Editions; applicable action and messaging licences may also be needed. This is Service Agent guidance, not universal eligibility for every Agentforce product.",
      "last_checked": "2026-10-24",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Service and employee support workflows",
          "source_url": "https://www.salesforce.com/agentforce/",
          "observed_at": "2026-10-12",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Agent builder and observability",
          "source_url": "https://www.salesforce.com/agentforce/",
          "observed_at": "2026-10-12",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Voice/phone, web and in-app channels with hand-off to humans",
          "source_url": "https://www.salesforce.com/agentforce/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Sales development: answering product questions and booking meetings",
          "source_url": "https://www.salesforce.com/agentforce/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "For Service Agents, Salesforce documents a dedicated Agent User rather than the logged-in admin or service rep. Explicitly grant required object, flow, prompt and data access.",
        "Service Agent builder/setup permissions and applicable permission-set licences are separate from end-user employee entitlements. Review the selected agent type and channel before deployment.",
        "A documented agent identity or permission set is not proof that every generated action is safe; review configured actions and output."
      ],
      "limitations": [
        "Flex Credits and Conversations are mutually exclusive within one org, according to the pricing FAQ.",
        "The $5 metered employee licence is not the $125 unmetered employee add-on; verify edition prerequisites and actual entitlements.",
        "Service Agent setup guidance does not establish prerequisites for every employee or industry agent.",
        "Digital Wallet alerts do not establish a universal hard budget stop. Overage follows the contracted rate.",
        "Vendor documentation only; no hands-on task quality, security or reliability test."
      ],
      "runtime": [
        "Enterprise platform"
      ],
      "privacy_security": null,
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": "Different consumption and employee-licensing routes; no universal allowance claimed",
        "usage_charges": "Flex Credit overage is billed at the contracted rate monthly in arrears; unused credits do not roll into later subscription terms.",
        "verified_at": "2026-10-24",
        "source_url": "https://www.salesforce.com/agentforce/pricing/",
        "note": "Salesforce lists Flex Credits at USD $500 per 100,000 credits (20 credits per standard action, 30 per Agentforce Voice action), employee add-ons at $125/user/month ($150 Industries), Max Edition from $550/user/month including 2.75M credits per org per year, a $5/user/month metered User License (requires Flex Credits) and a $50/user/month Coworker licence. A per-conversation model is listed for customer-facing agents but no per-conversation price is published on the current page."
      },
      "sources": [
        "https://www.salesforce.com/agentforce/",
        "https://www.salesforce.com/agentforce/pricing/",
        "https://help.salesforce.com/s/articleView?id=005317651&language=fi&type=1"
      ],
      "changelog": [
        {
          "date": "2026-10-22",
          "change": "Pricing re-verified against the live vendor pricing page and expanded: $0 Salesforce Foundations entry, Max Edition (add-on plus 2.75M Flex Credits/org/year), Agentforce Coworker User License (standalone per-user SKU, searches included, 30 MB indexing/user/month), Voice actions at 30 Flex Credits. Capabilities extended with voice/phone channels and sales development. No hands-on testing."
        },
        {
          "date": "2026-10-12",
          "change": "Added source-linked licence, consumption and Service Agent prerequisite detail, eight evidence rows, narrow setup and Salesforce's labelled order-status example. No total-cost or tested-output claim."
        },
        {
          "date": "2026-10-01",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-10-24",
          "change": "Pricing correction: the per-conversation price previously shown is not published on the current Salesforce pricing page or rate cards; removed the figure and added the pricing and licensing guide link."
        }
      ],
      "coverage_rows": [
        [
          "Charging model",
          "Org-level choice",
          "Salesforce says Flex Credits and conversation pricing cannot coexist in the same org. Switching requires replacing conversation SKUs, not simply picking a cheaper unit per task.",
          "https://www.salesforce.com/agentforce/pricing/"
        ],
        [
          "Customer-facing work",
          "Two consumption routes",
          "Conversations are positioned for external customer agents at USD $2 each; Flex Credits cover customer and employee use cases. A conversation is not the same unit as an action.",
          "https://www.salesforce.com/agentforce/pricing/"
        ],
        [
          "Action metering",
          "Standard vs voice",
          "The pricing page lists $500 per 100,000 Flex Credits and 20 credits per standard action. Voice actions use 30 credits; other usage/multipliers require the rate card.",
          "https://www.salesforce.com/agentforce/pricing/"
        ],
        [
          "Employee access",
          "Licence plus metered use",
          "Agentforce User License is listed at $5/user/month with limited CRM-object access and metered employee usage. Do not treat it as the unlimited employee add-on.",
          "https://www.salesforce.com/agentforce/pricing/"
        ],
        [
          "Employee flat fee",
          "Separate entitlement",
          "Agentforce add-ons are listed from $125/user/month with unmetered employee usage for licensed users. Cloud/industry edition prerequisites and included rights require the actual offer.",
          "https://www.salesforce.com/agentforce/pricing/"
        ],
        [
          "Free entry and editions",
          "Free start, bundle ceiling",
          "Salesforce markets a $0 Salesforce Foundations entry to start with Agentforce. Max Edition bundles an employee add-on plus 2.75M Flex Credits per org per year; an Agentforce Coworker User License is a standalone per-user SKU with searches included and 30 MB indexing per user per month. Actual entitlements require the current offer.",
          "https://www.salesforce.com/agentforce/pricing/"
        ],
        [
          "Service-agent prerequisites",
          "Use-case scoped",
          "Salesforce's Service Agent readiness guide calls for Lightning, Data Cloud, Einstein Generative AI/Trust Layer and Agentforce enablement, plus applicable builder/user/template/messaging licences. This is not a universal checklist for every Agentforce product.",
          "https://help.salesforce.com/s/articleView?id=005317651&language=fi&type=1"
        ],
        [
          "Agent identity",
          "Separate permissions",
          "The Service Agent guide describes a dedicated Agent User with explicitly granted object, flow, prompt and data access, not the logged-in admin's unrestricted identity.",
          "https://help.salesforce.com/s/articleView?id=005317651&language=fi&type=1"
        ],
        [
          "Overage and expiry",
          "Contract-dependent bill",
          "Unused Flex Credits do not roll into later subscription terms. Usage beyond entitlement is billed at the contracted rate monthly in arrears; Digital Wallet alerts are not a universal hard stop.",
          "https://www.salesforce.com/agentforce/pricing/"
        ]
      ],
      "setup_steps": [
        "Identify customer-facing or employee work and the exact Service/employee product before selecting a licence.",
        "Validate supported edition and prerequisites, and grant only the dedicated Agent User permissions needed.",
        "Choose the org's consumption model and review user add-ons, usage entitlement and overage terms together.",
        "Run one bounded low-risk task with reviewed output and record its actual actions and consumption before scaling."
      ],
      "price_tags": [
        "usage"
      ],
      "price_summary": "Flex Credits $500 per 100,000; add-ons $125/user/month; conversation price not published",
      "pricing_guide": {
        "href": "/guides/agentforce-pricing-and-licensing-explained/",
        "label": "Full Agentforce pricing and licensing guide"
      },
      "extra_faq": [
        {
          "question": "What does Salesforce's $120 example include?",
          "answer": "Salesforce's own order-status example uses 2 standard actions per request: 40 Flex Credits. At 20 requests/day for 30 days it lists 24,000 credits and $120 usage. This is its illustrative workload, not the complete monthly bill, minimum purchase, licence quote or a measured result. Prerequisites, add-ons, other consumption and tax remain separate.",
          "sources": [
            "https://www.salesforce.com/agentforce/pricing/"
          ]
        }
      ],
      "page_url": "https://agentcomparison.net/agents/agentforce/"
    },
    {
      "id": "manus-cue",
      "name": "Cue by Manus",
      "vendor": "Manus",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Manus describes Cue as a standalone app for a personal team of agents on mobile and desktop, where each agent has its own email, phone number, wallet and computer.",
      "one_liner": "A personal team of agents, each with its own email, phone number and wallet.",
      "best_fit": "People who want several agents with separate identities that hand work to each other, and who will review the final result.",
      "poor_fit": "Anyone who needs published spending limits, retention terms or a public price before connecting accounts; none was established in the source we read.",
      "decision_note": "Check what each agent's wallet and phone number can do, and what spending limits and approvals apply, before inviting it into your accounts. Access currently needs an invitation code.",
      "availability": "Manus's help center says Cue currently requires an invitation code, obtained through the Agents (Cue!) tab in the Manus sidebar.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Personal agents with their own email, phone number, wallet and computer",
          "source_url": "https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Invitation code required",
          "region_restrictions": null
        },
        {
          "name": "Group chats where several agents hand work to each other",
          "source_url": "https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Invitation code required",
          "region_restrictions": null
        },
        {
          "name": "Single personal assistant for errands, reminders and research",
          "source_url": "https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Invitation code required",
          "region_restrictions": null
        },
        {
          "name": "Simultaneous iMessage, Telegram, LINE and Slack Messenger connections",
          "source_url": "https://help.manus.im/en/articles/14178640-can-i-connect-multiple-messaging-agents-at-once-in-cue",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Invitation code required",
          "region_restrictions": null
        },
        {
          "name": "Scheduled Automations through Cue",
          "source_url": "https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Invitation code required",
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Manus says each agent has its own email, phone number, wallet and computer, and that you set direction and make the final call. Transaction limits were not detailed in the source we read."
      ],
      "limitations": [
        "Invitation code required at the time we checked",
        "Wallet limits, approval rules and data retention were not stated on the help page we read",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Mobile and desktop app built on the same infrastructure as Manus, per Manus"
      ],
      "privacy_security": "Cue-specific retention, training and security controls were not established by the two product help articles checked. General Manus policies or features are not treated as proof of Cue-specific controls. Confirm the applicable policy before connecting sensitive accounts.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-22",
        "source_url": "https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0",
        "note": "The Manus help article describes Cue but gives no price. Whether Cue uses Manus credits, and what wallet transactions cost, are not established. Unknown is not free.",
        "class": "see_vendor"
      },
      "sources": [
        "https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0",
        "https://help.manus.im/en/articles/14178640-can-i-connect-multiple-messaging-agents-at-once-in-cue"
      ],
      "changelog": [
        {
          "date": "2026-10-22",
          "change": "Re-verified both Manus help sources: invitation-code requirement, per-agent email/phone/wallet/computer identities, team channels with handoffs, Scheduled Automations and the four Messenger integrations (iMessage, Telegram, LINE, Slack) all still current. No pricing published; unknown is not free. No hands-on testing."
        },
        {
          "date": "2026-10-10",
          "change": "Expanded source-linked feature evidence and narrow setup; clarified access, permissions, costs and unresolved limits. Documentation research only, not a product test."
        },
        {
          "date": "2026-10-01",
          "change": "Added after Manus announced Manus 2.0 including Cue (vendor help article; trade press report dated 29 September 2026). Documentation research only."
        }
      ],
      "coverage_rows": [
        [
          "Agent identities",
          "Documented",
          "Each agent has its own email, phone number, wallet and computer. Payment provider, purchase limits and approval mechanics are not specified in this article.",
          "https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0"
        ],
        [
          "Team channels and handoffs",
          "Documented",
          "A channel can hold several agents with different jobs; agents pass work to one another. This is not proof of arbitrary public group-chat access.",
          "https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0"
        ],
        [
          "Recurring work",
          "Documented",
          "Create a Scheduled Automation through Cue by describing what should happen and when. Do not assume every general Manus automation or desktop feature belongs to Cue.",
          "https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0"
        ],
        [
          "Messaging connections",
          "Documented",
          "The Cue Messenger guide lists iMessage, Telegram, LINE and Slack; multiple connections may be enabled at once.",
          "https://help.manus.im/en/articles/14178640-can-i-connect-multiple-messaging-agents-at-once-in-cue"
        ],
        [
          "Access and signup",
          "Invitation required",
          "Open Agents (Cue!) in the Manus sidebar, scan the QR code to download the app, then enter the invitation code at sign-in. A code may also arrive by email.",
          "https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0"
        ],
        [
          "Price and wallet controls",
          "Unknown",
          "The checked Cue documentation does not establish a Cue-specific price, credit mapping, payment rail, spending cap or per-purchase approval policy. Unknown is not free.",
          "https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0"
        ]
      ],
      "setup_steps": [
        "Confirm invitation access before planning a workflow.",
        "Start with one research or reminder agent, not a wallet-bearing task.",
        "Connect a messaging platform under Agent Info > Integration > Messenger; review each authorization and toggle.",
        "For several agents, define the separate jobs and review what they hand off. Confirm payment controls separately before a purchase."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Price not verified; invitation code required",
      "meta_title": "Cue by Manus: what it is, invitation code access",
      "page_url": "https://agentcomparison.net/agents/manus-cue/"
    },
    {
      "id": "doordash-text",
      "name": "DoorDash text ordering",
      "vendor": "DoorDash",
      "kind": "product",
      "category": "food ordering assistant",
      "group": "errands-travel-food",
      "summary": "DoorDash describes a text-to-order assistant, built on its in-app Ask DoorDash, that finds restaurants and groceries, suggests a cart and checks out inside a text thread, using your past orders and preferences.",
      "one_liner": "A text-to-order assistant for restaurants and groceries on DoorDash.",
      "best_fit": "People who reorder the same food or groceries and want to do it by text without opening an app.",
      "poor_fit": "Anyone outside the US, or who needs published limits on automatic checkout and data use before connecting a payment method; none was established in the source we read.",
      "decision_note": "Check what it can order and pay for without confirmation, what it learns about you and how to turn proactive messages off. It is in a US-only beta behind a waitlist.",
      "availability": "DoorDash says US users can apply to try the beta through a waitlist at doordash.com/text (announced 30 September 2026).",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Order food or groceries by text, including \"order my usual\"",
          "source_url": "https://about.doordash.com/en-us/news/text-doordash",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "US beta waitlist",
          "region_restrictions": "US"
        },
        {
          "name": "Cart suggestions with photos of recommended dishes",
          "source_url": "https://about.doordash.com/en-us/news/text-doordash",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "US beta waitlist",
          "region_restrictions": "US"
        },
        {
          "name": "Group orders with mixed dietary preferences and quantities",
          "source_url": "https://about.doordash.com/en-us/news/text-doordash",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "US beta waitlist",
          "region_restrictions": "US"
        },
        {
          "name": "Optional proactive messages based on order habits",
          "source_url": "https://about.doordash.com/en-us/news/text-doordash",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "US beta waitlist",
          "region_restrictions": "US"
        }
      ],
      "permissions": [
        "Operates on the customer's existing DoorDash account: it knows past orders ('order my usual'), dietary preferences and modifiers from order history.",
        "Can build carts, suggest cheaper swaps, and manage checkout inside the text thread once the customer confirms the order.",
        "Vendor states it may message proactively (e.g. suggesting your usual coffee order) based on daily habits."
      ],
      "limitations": [
        "US beta, waitlist only at the time we checked",
        "Whether any fee applies was not stated in the announcement",
        "Approval rules beyond confirming the order in the text thread, and data retention, were not detailed",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Text messaging with DoorDash; the announcement refers to a text thread, TechCrunch reported Apple Messages"
      ],
      "privacy_security": "No privacy or data-use statement specific to text ordering is published in the launch announcement; ordering runs under the standard DoorDash account, so DoorDash's general privacy practices apply. Personalization ('order my usual', dietary preferences, proactive suggestions) is driven by order history per the announcement. Checked 22 October 2026.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-22",
        "source_url": "https://about.doordash.com/en-us/news/text-doordash",
        "note": "The DoorDash announcement gives no price for the text beta. Ordinary DoorDash order costs and fees presumably still apply, but the announcement does not say. Unknown is not free."
      },
      "sources": [
        "https://about.doordash.com/en-us/news/text-doordash"
      ],
      "changelog": [
        {
          "date": "2026-10-22",
          "change": "Field-level audit from the 30 Sep 2026 launch announcement: US beta via waitlist (doordash.com/text); text-to-cart with in-thread checkout confirmation; personalization from order history; suggested cheaper swaps; no assistant fee mentioned (standard order costs). Privacy: no feature-specific statement; standard DoorDash account practices apply. Coverage rows, permissions, setup steps filled. No hands-on testing."
        },
        {
          "date": "2026-10-22",
          "change": "Re-verified the DoorDash announcement: text ordering with usual picks, cart suggestions with dish photos, group orders with mixed dietary preferences all still described; no price published (unknown is not free). No hands-on testing."
        },
        {
          "date": "2026-10-01",
          "change": "Added after DoorDash announced text ordering on 30 September 2026 (DoorDash newsroom; TechCrunch reported it the same day). Documentation research only."
        }
      ],
      "coverage_rows": [
        [
          "Access",
          "US beta waitlist",
          "Launch announcement (30 September 2026): US users apply to try the beta via a waitlist at doordash.com/text. Not generally available.",
          "https://about.doordash.com/en-us/news/text-doordash"
        ],
        [
          "Ordering model",
          "Text-to-cart with in-thread checkout",
          "Text what you want; DoorDash searches local spots, builds and texts a suggested cart with photos, and checkout is confirmed inside the text thread.",
          "https://about.doordash.com/en-us/news/text-doordash"
        ],
        [
          "Pricing",
          "Standard order costs",
          "No subscription or assistant fee is mentioned in the announcement; normal DoorDash order pricing, fees and DashPass terms apply.",
          "https://about.doordash.com/en-us/news/text-doordash"
        ],
        [
          "Savings features",
          "Suggested swaps",
          "Scans the store in real time for cheaper equivalent swaps and totals potential savings before checkout; accept or decline per swap.",
          "https://about.doordash.com/en-us/news/text-doordash"
        ]
      ],
      "setup_steps": [
        "Join the US beta waitlist at doordash.com/text from your DoorDash account.",
        "When access arrives, start with a simple reorder ('order my usual') and check the suggested cart and fees before confirming in-thread.",
        "Review any proactive-message settings if you do not want habit-based suggestions."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Price not verified; US beta waitlist",
      "page_url": "https://agentcomparison.net/agents/doordash-text/"
    },
    {
      "id": "mindtrip",
      "name": "Mindtrip Personal Travel Assistant",
      "vendor": "Mindtrip",
      "kind": "product",
      "category": "travel planning assistant",
      "group": "errands-travel-food",
      "summary": "Mindtrip describes a travel assistant that learns a traveler's preferences, remembers them between trips and can run agents that watch for flights, hotel prices and local ideas on the traveler's behalf.",
      "one_liner": "A travel planner that remembers your preferences and watches flights and hotels.",
      "best_fit": "Travellers who plan several trips a year and want preferences remembered, with monitoring of prices and options.",
      "poor_fit": "Anyone who needs published pricing, booking terms and data-retention details before sharing travel and loyalty information; none was in the source we read.",
      "decision_note": "Check what it can book or change on its own, how preferences and loyalty-programme details are stored, and what the agents do while you are away. The source is the vendor press release only.",
      "availability": "Mindtrip announced the launch on 30 September 2026 and points to Mindtrip.ai. The release does not state plans, regions or an access limit.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Remembers traveler preferences, companions, budget and loyalty programmes across trips",
          "source_url": "https://www.prnewswire.com/news-releases/mindtrip-launches-personal-travel-assistant-that-knows-the-traveler-not-just-the-trip-302893318.html",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "low",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Background agents that watch flights, hotel prices and local ideas",
          "source_url": "https://www.prnewswire.com/news-releases/mindtrip-launches-personal-travel-assistant-that-knows-the-traveler-not-just-the-trip-302893318.html",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "low",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Itineraries, destination briefings and packing suggestions",
          "source_url": "https://www.prnewswire.com/news-releases/mindtrip-launches-personal-travel-assistant-that-knows-the-traveler-not-just-the-trip-302893318.html",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "low",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Booking through Mindtrip",
          "source_url": "https://www.prnewswire.com/news-releases/mindtrip-launches-personal-travel-assistant-that-knows-the-traveler-not-just-the-trip-302893318.html",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "low",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Account sign-in can use third-party services such as Google; Mindtrip receives profile information permitted by that service's settings.",
        "Optional voice input is transcribed and retained as conversation history for the duration of the active account unless deleted earlier.",
        "Payments are processed by third-party payment processors when booking through the service.",
        "The service can also be embedded on third-party sites (hotels, DMOs), which are governed by their own privacy policies."
      ],
      "limitations": [
        "Based on a vendor press release; no help-centre or pricing page read",
        "Price, regions and booking terms were not stated",
        "Mindtrip says it is independent of travel suppliers; we have not verified how it earns revenue",
        "Performance figures in the release (18,000 beta travelers, 3x more likely to create a trip) are Mindtrip's own and unaudited",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Mindtrip platform; agents run on the vendor side, per the release"
      ],
      "privacy_security": "Mindtrip's privacy policy states it does not use identifiable personal data to train or fine-tune AI models, that third-party AI providers are contractually prohibited from training or fine-tuning on user data, and that it does not use personal data to develop, improve or train generalized AI/ML models; it also says it does not allow humans to read it except in limited cases. Voice transcripts are retained as conversation history for the life of the active account unless deleted earlier; access, correction or deletion can be requested. Checked 22 October 2026.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-22",
        "source_url": "https://mindtrip.ai/business/packages",
        "note": "No verified consumer price list. The consumer app store listing markets the app as free with no subscription mentioned; Mindtrip for Business lists Basic, Premium and Enterprise packages (Enterprise is 'contact us'; package prices not shown in the page text we could read). The launch press release gives no price. Unknown is not free."
      },
      "sources": [
        "https://www.prnewswire.com/news-releases/mindtrip-launches-personal-travel-assistant-that-knows-the-traveler-not-just-the-trip-302893318.html",
        "https://mindtrip.ai/business/packages",
        "https://apps.apple.com/us/app/mindtrip-ai-travel-companion/id6503107567"
      ],
      "changelog": [
        {
          "date": "2026-10-22",
          "change": "Field-level audit: privacy policy captured - no training/fine-tuning on identifiable personal data, AI sub-processors contractually barred from training, voice transcripts retained for account life unless deleted earlier. Consumer app marketed free (App Store); no consumer price list; Business Basic/Premium/Enterprise confirmed. Coverage rows, permissions, setup steps, privacy fields filled. No hands-on testing."
        },
        {
          "date": "2026-10-22",
          "change": "Pricing picture expanded: consumer App Store listing markets the app as free; Mindtrip for Business lists Basic/Premium/Enterprise packages with message/indexing/guide allowances (Enterprise 'contact us'; no dollar amounts in readable page text). No verified consumer price list; unknown is not free. No hands-on testing."
        },
        {
          "date": "2026-10-01",
          "change": "Added after Mindtrip's press release of 30 September 2026. Vendor press release only; documentation research, not tested."
        }
      ],
      "coverage_rows": [
        [
          "Consumer pricing",
          "Free per App Store listing",
          "The consumer App Store listing markets the app as free; no consumer price list published. Business packages (Basic/Premium/Enterprise) are separately listed.",
          "https://apps.apple.com/us/app/mindtrip-ai-travel-companion/id6737068622"
        ],
        [
          "Business packages",
          "Basic / Premium / Enterprise",
          "Mindtrip for Business lists Basic, Premium and Enterprise packages with message, indexing and guide allowances; Enterprise is 'contact us'.",
          "https://mindtrip.ai/business/packages"
        ],
        [
          "Training use",
          "No training on identifiable data",
          "Privacy policy: no training or fine-tuning on identifiable personal data; third-party AI providers contractually prohibited from training; no human reading except limited cases.",
          "https://mindtrip.ai/privacy-policy"
        ],
        [
          "Voice retention",
          "Life of account",
          "Voice transcripts retained as conversation history for the active account's duration unless deleted earlier; deletion on request.",
          "https://mindtrip.ai/privacy-policy"
        ]
      ],
      "setup_steps": [
        "Install the consumer app (marketed free) or start on the web, and sign in - Google sign-in shares the profile fields that service permits.",
        "Plan one low-stakes trip first; review what preference information you share before enabling voice input.",
        "If voice transcripts concern you, note they persist for the account's life unless you delete them earlier."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Price not stated in launch release",
      "page_url": "https://agentcomparison.net/agents/mindtrip/"
    },
    {
      "id": "wabi",
      "name": "Wabi",
      "vendor": "Wabi, Inc.",
      "kind": "product",
      "category": "messenger-style personal agent",
      "group": "personal-assistants",
      "summary": "Wabi 3.0 describes an AI messenger where you ask for plans, errands or small apps and it does the work, with friends or family in the same chat. Wabi says it can build a plan, tracker, map, list or game inside the chat and can raise things worth handling when you connect accounts.",
      "one_liner": "A messenger-style agent for plans, errands and small apps, shareable with friends.",
      "best_fit": "People who want to coordinate plans and errands with others in one chat and are comfortable with an invite-only early product.",
      "poor_fit": "Anyone who needs published pricing, approval rules or reliability data first, or who is not on iPhone.",
      "decision_note": "Now listed publicly on the App Store as 'Wabi: AI Assistant' (version 3.0, 76 ratings at 4.4 when checked 22 October 2026). Read what you connect, what needs your approval, and what the credits and subscription cost before relying on it. Wabi is not in our text-message assistants table under the messaging-service inclusion rule.",
      "availability": "The US App Store lists it for iPhone only, 13+. TechCrunch (29 Sep 2026) reports Wabi 2.0 is available by invite code, handed out on X. Android listing exists on Google Play; we did not read it.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Handle errands and plans from a chat request, such as booking an appointment or planning dinner (as described)",
          "source_url": "https://apps.apple.com/us/app/wabi-ai-messenger/id6747768928",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Invite code required per TechCrunch",
          "region_restrictions": "App Store listing is the US storefront; iPhone only"
        },
        {
          "name": "Group chats where everyone sees and moves the same plan",
          "source_url": "https://apps.apple.com/us/app/wabi-ai-messenger/id6747768928",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Invite code required per TechCrunch",
          "region_restrictions": "App Store listing is the US storefront; iPhone only"
        },
        {
          "name": "Create a plan, tracker, map, list or game inside the chat",
          "source_url": "https://apps.apple.com/us/app/wabi-ai-messenger/id6747768928",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Invite code required per TechCrunch",
          "region_restrictions": "App Store listing is the US storefront; iPhone only"
        },
        {
          "name": "Raise deadlines, bills or forgotten plans when you connect accounts",
          "source_url": "https://apps.apple.com/us/app/wabi-ai-messenger/id6747768928",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Invite code required per TechCrunch",
          "region_restrictions": "App Store listing is the US storefront; iPhone only"
        }
      ],
      "permissions": [
        "Device permissions are per-permission and revocable in OS settings: contacts, photo library/camera, microphone, Reminders, and on Mac the screen of an app you ask about plus Desktop/Documents/Downloads, Notes and Messages - Messages is read only when a task needs it and Wabi sends only messages you approve.",
        "Background location is optional and off unless you enable it; reports are rounded to about 100 metres and are not shared with advertising partners.",
        "Connected accounts (mail, calendar, Slack, Notion, GitHub and others) connect individually and can be revoked per connector (e.g. Profile > Connectors > Slack > Disconnect); Slack is read only when you ask, not continuously.",
        "Payment connections (Stripe/Plaid) are opt-in and scoped to tasks you request; Wabi receives merchant receipts for things it does on your behalf."
      ],
      "limitations": [
        "Invite-only per TechCrunch at the time we checked",
        "Pricing and credit amounts are not published on the pages we read",
        "Only 8 App Store ratings (3.6 average) when checked; one review reports it stopped working after a day",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "iPhone app (iOS)",
        "Group chats with other people"
      ],
      "privacy_security": "Privacy policy (updated 19 October 2026) and terms (updated 20 October 2026), both read 22 October 2026. Connected-account content (Gmail, iCloud, Outlook, Google Calendar, Drive, Slack, Notion, Granola, GitHub, Postedly, Telegram, WhatsApp) is used only to provide the features you ask for and is NOT used to train AI models - Wabi's own or its providers' - nor for advertising or sale; Google data is handled under Limited Use, and humans do not read Google data except with consent or for security/legal. Requests are processed by Google, Anthropic and OpenAI under agreements that prohibit training on your data and limit retention. Voice/video calls are transcribed for the assistant; call audio and video are not stored. Health data is never used for advertising and is deleted within 30 days of disconnecting. The flip side: a large advertising stack is disclosed (Meta Pixel/Conversions API/Custom Audiences, TikTok, Snap and others, with hashed emails and device identifiers shared) plus analytics vendors (Amplitude, AppsFlyer, Datadog, Firebase, Loops), with opt-outs via App Tracking Transparency, device settings and privacy@wabi.ai. Slack content used in a conversation is kept with that conversation until you delete it or your account. Vendor policy, not independently audited.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-22",
        "source_url": "https://apps.apple.com/us/app/wabi-ai-assistant/id6747768928",
        "note": "The App Store lists the app download as Free. Wabi's terms say a free version and paid subscriptions may carry credits, that subscriptions are bought through Apple In-App Purchase, and that fees are in US dollars, but no plan prices or credit amounts were on the pages we read. A free download is not the same as free use."
      },
      "sources": [
        "https://apps.apple.com/us/app/wabi-ai-assistant/id6747768928",
        "https://wabi.ai/",
        "https://wabi.ai/terms",
        "https://techcrunch.com/2026/09/29/ai-powered-app-maker-wabi-pivots-to-a-messaging-experience/",
        "https://wabi.ai/privacy"
      ],
      "changelog": [
        {
          "date": "2026-10-02",
          "change": "Added at Heni's request after the App Store listing; checked against the App Store page, wabi.ai terms and TechCrunch (29 Sep 2026). Documentation research only."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: privacy policy (updated 19 October 2026) and terms (20 October 2026) read for the first time - connected-account content excluded from AI training, Google/Anthropic/OpenAI under no-training agreements, calls transcribed without storing audio/video, health data deleted within 30 days of disconnect, alongside a large disclosed advertising stack with opt-outs. Permissions field rewritten with the real per-permission model."
        }
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Free download; plan prices not published on pages read",
      "page_url": "https://agentcomparison.net/agents/wabi/"
    },
    {
      "id": "szn",
      "name": "szn",
      "vendor": "szn",
      "kind": "product",
      "category": "messaging-first personal agent",
      "group": "personal-assistants",
      "summary": "szn describes an AI assistant you reach in one message thread by text, voice note, photo or forwarded email. Its site says it makes calls, sends emails and follows up on your behalf, and lists tasks such as finding a time, planning a trip and watching something for you.",
      "one_liner": "A one-thread assistant for calls, email and tasks by text, voice note or photo.",
      "best_fit": "People who want to hand off phone calls, email follow-ups and bookings from one thread and are willing to wait for published terms.",
      "poor_fit": "Anyone who needs a published price, approval rules, supported countries or reliability data before connecting accounts or paying for purchases.",
      "decision_note": "Announced in late September 2026, with press saying an October launch. Price, regions, approval rules and exact launch date were not disclosed in what we read. Treat it as pre-launch until szn publishes them.",
      "availability": "Press (ithinkdiff, 27 Sep 2026; MacObserver) says szn is expected to launch in October through iMessage, with a dedicated app later. We did not find a public sign-up or pricing page on theszn.ai.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Reach it in one thread by text, voice note, photo or forwarded email",
          "source_url": "https://theszn.ai/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Not disclosed; pre-launch per press",
          "region_restrictions": "Not disclosed. The site says it has served users in 16 countries, unverified."
        },
        {
          "name": "Make calls and send emails and follow-ups on your behalf (as described)",
          "source_url": "https://theszn.ai/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Not disclosed; pre-launch per press",
          "region_restrictions": "Not disclosed. The site says it has served users in 16 countries, unverified."
        },
        {
          "name": "Plan trips, schedule and keep an eye on things for you (as described)",
          "source_url": "https://theszn.ai/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Not disclosed; pre-launch per press",
          "region_restrictions": "Not disclosed. The site says it has served users in 16 countries, unverified."
        }
      ],
      "permissions": [
        "One thread for everything: text, voice note, photo or a forwarded email over iMessage.",
        "Takes real-world actions on the user's behalf: 'calls get made, emails get sent, follow-ups happen' - the user steps back in for decisions (vendor description).",
        "May share relevant information with third parties involved in completing a requested task (per privacy policy).",
        "Sign-in via Google or Apple identity providers."
      ],
      "limitations": [
        "Pre-launch per press; no exact date",
        "Price, supported countries and approval rules not disclosed in what we read",
        "Press reports it can browse, make purchases and contact vendors; spending and approval limits are unstated",
        "Described by the vendor and press; not tested by AgentComparison"
      ],
      "runtime": [
        "iMessage thread (per press)",
        "Vendor-hosted"
      ],
      "privacy_security": "SZN's privacy policy says that to complete requested tasks the service may share relevant information with third parties involved in the task, and that sign-in via Google or Apple passes basic profile information. It describes CCPA rights for California users. The marketing line is 'private at every step', but the public policy checked 23 October 2026 contains no AI-training, retention-period or deletion-mechanics statement; treat those as unpublished.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-21",
        "source_url": "https://theszn.ai/",
        "note": "No price on theszn.ai or in the press we read (MacObserver says pricing, countries and launch date were not disclosed). Unknown is not free.",
        "class": "see_vendor"
      },
      "sources": [
        "https://theszn.ai/",
        "https://theszn.ai/privacy",
        "https://www.ithinkdiff.com/imessage-ai-agent-szn-former-apple-engineer/",
        "https://www.macobserver.com/news/szn-imessage-ai-agent-former-siri-engineer/"
      ],
      "changelog": [
        {
          "date": "2026-10-23",
          "change": "Field-level audit from live site + privacy policy: one-thread task intake (text/voice/photo/forwarded email); vendor says calls/emails/follow-ups happen with user stepping in for decisions; policy allows sharing relevant info with task-involved third parties; Google/Apple sign-in; CCPA rights; pricing unpublished; training/retention unpublished (recorded unknown). Coverage rows, permissions, setup steps, privacy fields filled. No hands-on testing."
        },
        {
          "date": "2026-10-02",
          "change": "Added from press launch coverage (27-28 Sep 2026) and theszn.ai. Documentation research only."
        },
        {
          "date": "2026-10-21",
          "change": "Re-checked vendor site: no pricing published, claims unchanged."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Not published",
          "No price on theszn.ai or in the press coverage read (MacObserver notes pricing, countries and launch date were undisclosed). Unknown is not free.",
          "https://theszn.ai/"
        ],
        [
          "Action model",
          "Acts, then reports",
          "Vendor: send a task by text/voice/photo/forwarded email; szn makes calls, sends emails and does follow-ups; the user steps back in for decisions.",
          "https://theszn.ai/"
        ],
        [
          "Data sharing",
          "Task-third-parties",
          "Privacy policy: relevant information may be shared with third parties involved in completing the requested task.",
          "https://theszn.ai/privacy"
        ],
        [
          "Training/retention",
          "Unpublished",
          "No AI-training or retention statement found in the public policy as of 23 October 2026; recorded as unknown.",
          "https://theszn.ai/privacy"
        ]
      ],
      "setup_steps": [
        "Start with one low-stakes errand by text and watch what it shares with third parties to complete it.",
        "Note that pricing is unpublished - confirm the cost model inside the product before relying on it.",
        "Ask in-thread how your data is retained before forwarding sensitive emails or documents."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Not disclosed; pre-launch",
      "page_url": "https://agentcomparison.net/agents/szn/"
    },
    {
      "id": "underdog",
      "name": "Underdog",
      "vendor": "Conway Research (Underdog)",
      "kind": "product",
      "category": "on-device personal agent",
      "group": "personal-assistants",
      "summary": "Underdog describes a personal AI app for email, calendar, meeting notes, dictation and a browser agent that runs its own models on your Mac, with content not sent to a cloud AI service. Its site lists expense tracking and tax preparation as coming soon.",
      "one_liner": "An on-device Mac assistant for email, calendar, meeting notes and a browser agent.",
      "best_fit": "Mac users who want email, calendar, notes and dictation help and care most that AI processing stays on their computer.",
      "poor_fit": "Anyone not on an Apple silicon Mac, or who needs independent audit results, approval-control details and reliability data before connecting email and calendar.",
      "decision_note": "Launched by Conway Research; a16z announced it led the company's first funding round on 2 Oct 2026. The vendor says it is free and local. Its own pages disagree on access (invite-only beta versus 'today, if you have an Apple silicon Mac'), and its audits are partly still under way. Underdog is not in our text-message assistants table: its documented interface is an on-device Mac app (email, calendar, meeting notes, dictation, browser agent); no two-way messaging channel is documented.",
      "availability": "Invite-only beta. Mac, iPhone and iPad via the App Store; Windows in beta; Linux and Android announced.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Email, calendar, meeting notes and dictation on your Mac",
          "source_url": "https://underdog.ai/",
          "observed_at": "2026-10-02",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Invite-only beta per site header; free per pricing page",
          "region_restrictions": "Apple silicon Mac; other platforms listed as coming"
        },
        {
          "name": "Browser agent and 'sensitive tasks' (as listed on the site)",
          "source_url": "https://underdog.ai/",
          "observed_at": "2026-10-02",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Invite-only beta per site header; free per pricing page",
          "region_restrictions": "Apple silicon Mac; other platforms listed as coming"
        },
        {
          "name": "Local models; memory you can inspect and delete in Settings",
          "source_url": "https://underdog.ai/",
          "observed_at": "2026-10-02",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Invite-only beta per site header; free per pricing page",
          "region_restrictions": "Apple silicon Mac; other platforms listed as coming"
        },
        {
          "name": "Expense tracking and tax preparation (listed as coming soon)",
          "source_url": "https://underdog.ai/",
          "observed_at": "2026-10-02",
          "evidence_type": "vendor_documentation",
          "confidence": "low",
          "plan_restrictions": "Invite-only beta per site header; free per pricing page",
          "region_restrictions": "Apple silicon Mac; other platforms listed as coming"
        }
      ],
      "permissions": [
        "Sensitive tasks (doctor's appointments, prescription refills, tax paperwork): it finds what you need and 'gets the next step ready for your approval'.",
        "Handles email, calendar, meeting notes, dictation and errands on-device.",
        "When asked to buy something, it picks an option and pays with a one-time card created for that order; the vendor says it takes a small margin.",
        "Strips remote images, scripts and tracking pixels from incoming mail on sight."
      ],
      "limitations": [
        "Access wording conflicts across the vendor's own pages",
        "Audits by outside researchers are listed as under way, with no results yet",
        "The vendor says its models may not be the biggest; no independent quality test was run",
        "Benchmark and speed claims (including the Husky engine post) are the vendor's own",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Mac app (Apple silicon), local models",
        "iPhone and iPad app per privacy policy; FAQ says coming"
      ],
      "privacy_security": "Underdog's core claim is architectural: models run locally on your Mac (Woof 4B, Saluki 27B, Underdog 27B built on Qwen 3.8 27B), so content is not sent to a cloud AI service for processing. Mail syncs directly between your computer and your provider. Saved memories stay on-device and can be viewed, edited or deleted in Settings; each account has its own vault with the key stored on the device. 'Naval mode' cuts all outbound calls. The security page states it passed the Leviathan Security Group App Defense Alliance assessment (all 20 requirements) and that Google approved its OAuth access to Gmail and Calendar; both are vendor-stated and were not independently verified. Training use: no cloud training exposure is possible under the local-processing claim, but no separate training policy document was found. Checked 23 October 2026.",
      "pricing": {
        "amount": 0,
        "currency": "USD",
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-23",
        "source_url": "https://underdog.ai/pricing",
        "note": "Free. Homepage FAQ (checked 23 October 2026): the assistant is free; when it buys something for you it pays with a one-time card made for that order and the vendor 'makes a tiny bit of it'. No paid plan is listed."
      },
      "sources": [
        "https://underdog.ai/",
        "https://underdog.ai/pricing",
        "https://underdog.ai/security",
        "https://underdog.ai/privacy",
        "https://husky.underdog.ai/",
        "https://a16z.com/announcement/investing-in-conway-the-creator-of-underdog/",
        "https://cryptobriefing.com/conway-research-underdog-private-ai-funding/"
      ],
      "changelog": [
        {
          "date": "2026-10-02",
          "change": "Added at Heni's request. Checked underdog.ai home, pricing, security and privacy pages, the Husky engine post, and a16z's 2 Oct 2026 funding post. Documentation research only."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Free",
          "FAQ: 'It's free? What's the catch?' - the assistant is free; when it buys something for you it pays with a one-time card and the vendor 'makes a tiny bit of it'.",
          "https://underdog.ai/"
        ],
        [
          "Local processing",
          "100% local",
          "Summaries, drafts and answers come from a model running on your computer (Woof 4B, Saluki 27B, Underdog 27B built on Qwen 3.8 27B); content is not sent to a cloud AI service.",
          "https://underdog.ai/"
        ],
        [
          "Memory control",
          "On-device, inspectable",
          "Saved memories stay on your computer; see, edit or delete them in Settings. Each account gets its own vault with the key locked on the device.",
          "https://underdog.ai/"
        ],
        [
          "Offline mode",
          "Naval mode",
          "One switch and Underdog 'never calls home'. Mail syncs directly between your computer and your email provider.",
          "https://underdog.ai/"
        ],
        [
          "Security claims",
          "Vendor-stated",
          "Security page states it passed the Leviathan Security Group App Defense Alliance assessment (all 20 requirements) and that Google approved its OAuth access to Gmail and Calendar. Vendor-stated; not independently verified.",
          "https://underdog.ai/security"
        ]
      ],
      "setup_steps": [
        "Request access - the beta is invite-only.",
        "Install from the App Store on Mac, iPhone or iPad; a Windows beta is listed, with Linux and Android 'on the way'.",
        "Connect email and calendar; the first-run downloads a local model (Saluki 27B needs about 8 GB, fits a 16 GB laptop).",
        "Review saved memories and the vault in Settings; enable Naval mode if you want zero outbound calls."
      ],
      "price_tags": [
        "free"
      ],
      "price_summary": "Free, per the vendor pricing page (no paid plan listed)",
      "page_url": "https://agentcomparison.net/agents/underdog/"
    },
    {
      "id": "case",
      "name": "Case",
      "vendor": "Case (case-computers)",
      "kind": "product",
      "category": "persistent computer for AI agents",
      "group": "builders",
      "summary": "Case sells access to a persistent computer for an AI agent: it opens apps, does the work and sends you the result, keeping state between sessions. Since October 2026 the site says an AI is built in and connecting your own is optional - earlier copy (3 October 2026) described a machine you drive over MCP from tools like Claude Code or Cursor with your own model and key, and that MCP positioning no longer appears. Case is open source for self-hosting; the paid plan is a hosted computer with Case set up. A separate 'AI cofounder' product for founder marketing is in development.",
      "one_liner": "A persistent computer with a built-in AI agent; open source, hosted plan optional.",
      "best_fit": "Developers who want an agent to keep logged-in sessions and files between runs, for jobs such as scheduled browser tasks, and who are comfortable running or paying for the machine and supplying their own model key.",
      "poor_fit": "Anyone who wants a finished assistant, or who needs a published privacy policy, terms, security audit or support commitments before storing logins on a machine an agent drives.",
      "decision_note": "Developer infrastructure, not a ready-made assistant. Case previously listed one plan at $29 a month with your own LLM key, but the price is no longer shown on its homepage (checked 22 October 2026), so confirm cost with the vendor. Its code is on GitHub, but the repository licence reads 'Other' and the project is small.",
      "availability": "Hosted at case.computer or self-hosted with Docker from the GitHub repository. We did not find a public privacy policy or terms page at case.computer/privacy or /terms.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "A persistent desktop whose logins, cookies and files survive sleep (as described)",
          "source_url": "https://case.computer/",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Hosted plan $29/month; self-hosting possible",
          "region_restrictions": "Not stated"
        },
        {
          "name": "MCP tools such as computer_create, computer_login, computer_sleep and computer_navigate, plus screenshot, exec and files",
          "source_url": "https://case.computer/",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Hosted plan $29/month; self-hosting possible",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Human handoff for 2FA, captchas and approvals by phone link",
          "source_url": "https://case.computer/",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Hosted plan $29/month; self-hosting possible",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Saved skills and scheduled recurring runs on the machine",
          "source_url": "https://github.com/case-computers/case",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Hosted plan $29/month; self-hosting possible",
          "region_restrictions": "Not stated"
        },
        {
          "name": "AI built in (connecting your own model optional); open source with self-hosting; hosted plan includes a ready computer",
          "source_url": "https://case.computer/computer/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "'AI cofounder' product for founder marketing in development (drafts outreach, spots churn risks, suggests next moves)",
          "source_url": "https://case.computer/cofounder/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "The terms make you solely responsible for actions the agent executes on the computer through your prompts, automations or approvals, and Case disclaims liability for them.",
        "Earlier site copy described an encrypted vault for logins and human handoff by phone link for 2FA, captchas and approvals; neither appears on the current site, so treat those controls as unconfirmed.",
        "No granular approval regime is described in the current terms or pages; the product is explicitly beta, provided as-is."
      ],
      "limitations": [
        "No privacy policy or terms page found on the site",
        "Licence shown as 'Other' on GitHub; read it before commercial use",
        "Small early project: 3 stars, one listed contributor at the time we checked",
        "You pay your own LLM costs on top of the $29 plan",
        "Described by the vendor; not tested by AgentComparison",
        "MCP-tool copy and the encrypted-vault/phone-approval claims from the 3 October 2026 site no longer appear on the current site; current positioning says an AI is built in."
      ],
      "runtime": [
        "Hosted desktop or self-hosted Docker",
        "MCP server for Claude Code, Codex, Cursor or any agent loop"
      ],
      "privacy_security": "Privacy policy (effective 24 March 2026, read 22 October 2026) is short and local-first: Case collects only your login email (via Supabase) and chat session metadata (timestamps, session IDs, operational metadata). Screen data, agent execution traces, on-device memory and workflow artifacts, and your API keys and provider credentials are intended to stay on your machine and are not transmitted to Case servers by default. Model inference goes to whichever LLM provider you configure. Case says it runs no third-party analytics and does not sell personal data. The older claim about an encrypted login vault no longer appears on the current site. Vendor policy, not independently audited.",
      "pricing": {
        "amount": 8,
        "currency": "USD",
        "billing_period": "first month, then $15 per month",
        "allowances": null,
        "usage_charges": "Optional: connect your own LLM provider, in which case model usage is billed by your provider",
        "verified_at": "2026-10-22",
        "source_url": "https://case.computer/terms.html",
        "note": "Terms of Use and the /computer/ FAQ (both read 22 October 2026): $8 for the first month, then $15/month, renewing automatically; cancel future renewal any time; full refund if you email within 7 days of the first payment. Self-hosting the open-source code is free apart from your own model costs.",
        "class": "subscription"
      },
      "sources": [
        "https://case.computer/",
        "https://github.com/case-computers/case",
        "https://case.computer/terms.html",
        "https://case.computer/privacy.html",
        "https://case.computer/computer/",
        "https://case.computer/cofounder/"
      ],
      "changelog": [
        {
          "date": "2026-10-03",
          "change": "Added at Heni's request. Checked case.computer and the GitHub repository. Documentation research only."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: pricing corrected from stale $29/month (unconfirmed) to $8 first month then $15/month per terms and FAQ; local-first privacy policy read (only email + session metadata leave the device); product repositioning documented - AI now built in, MCP bring-your-own-model copy no longer on site; vault/phone-approval claims marked unconfirmed; cofounder product noted as in development."
        }
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Price no longer published on the homepage (last verified $29/month, 3 Oct 2026)",
      "page_url": "https://agentcomparison.net/agents/case/"
    },
    {
      "id": "cursor",
      "name": "Cursor cloud agents",
      "vendor": "Anysphere (Cursor)",
      "kind": "product",
      "category": "coding agent (editor and cloud agents)",
      "group": "coding-agents",
      "summary": "Cursor is an AI code editor that also runs cloud agents: isolated virtual machines that clone your repository, work on a separate branch, test the change and push it for review. Agents can be started from the editor, web, mobile, Slack, GitHub, Linear or the API.",
      "one_liner": "An AI code editor whose cloud agents work on branches and open pull requests.",
      "best_fit": "Developers who want background work on repository branches and can review diffs, tests and artifacts before a human merges the draft pull request. Configure restricted egress and minimum repository access first.",
      "poor_fit": "Sensitive repositories without reviewed cloud-storage and network policies, workflows needing approval for every command, or users who assume a fixed subscription eliminates on-demand charges. Self-hosted tools still use Cursor's cloud agent loop.",
      "decision_note": "Use cloud agents for bounded repository tasks that you can review as draft pull requests. They auto-run terminal commands, unlike local approval modes, and internet access is on by default. Before starting, restrict repositories and outbound domains, choose secret types deliberately, verify Privacy Mode, and check usage budgets. Privacy Mode does not mean the run leaves no stored transcript or repository snapshot.",
      "availability": "Cursor's pricing feature table (re-read 28 October 2026) includes cloud agents on Premium ($30/month) and above on Individual plans, and on Teams; Free and Plus are excluded. Enabling managed cloud agents requires admin setup on Cursor and the Git provider, followed by each triggering user connecting their own Git account. Repository access is limited to the repositories authorized through the app and the triggering user's existing access. Hobby is free but is not listed as including cloud agents.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Managed VM repository work with draft PR and artifact handoff",
          "source_url": "https://cursor.com/docs/cloud-agent/security",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Paid cloud-agent access; some organization controls are plan-specific",
          "region_restrictions": "Not verified"
        },
        {
          "name": "Managed-cloud egress modes with user, environment and team policy",
          "source_url": "https://cursor.com/docs/cloud-agent/security-network",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Paid cloud-agent access; some organization controls are plan-specific",
          "region_restrictions": "Not verified"
        },
        {
          "name": "Environment Variable, Runtime Secret and Build Secret scoping",
          "source_url": "https://cursor.com/docs/cloud-agent/security-network",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Paid cloud-agent access; some organization controls are plan-specific",
          "region_restrictions": "Not verified"
        },
        {
          "name": "Managed or self-hosted tool execution with the agent loop in Cursor cloud",
          "source_url": "https://cursor.com/docs/cloud-agent/choose-runtime",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Paid cloud-agent access; some organization controls are plan-specific",
          "region_restrictions": "Not verified"
        }
      ],
      "permissions": [
        "An admin installs the Git provider app and selects repositories; each user connects their own Git identity. Cursor documents that the agent cannot reach repositories the triggering user could not already reach.",
        "Cloud agents run terminal commands without per-action approval and open draft pull requests for human review before merging. Do not apply the local Auto-review, Allowlist or Run Everything settings as cloud-agent gates.",
        "Internet access is on by default. Managed cloud users, saved environments and teams can choose Allow all, Default + allowlist or Allowlist only; admins can lock policy. Scope exact destinations, not broad storage-domain wildcards.",
        "Environment Variable secrets are visible to the agent. Runtime Secrets are redacted from tool results, transcripts and commits, but still exist as environment variables and are visible to people using the environment Terminal. Build Secrets are available only during Docker builds.",
        "Secret redaction is not credential revocation. Remove stored secrets/tokens when finished, and check transcript and snapshot retention separately. Network controls for managed agents do not govern self-hosted workers."
      ],
      "limitations": [
        "Paid cloud-agent access is distinct from Hobby's limited local Agent requests; removed the Free cloud-agent filter tag",
        "No per-command approval: cloud agents auto-run terminal work, with prompt-injection and exfiltration risks described by Cursor",
        "Default internet access can be restricted; managed-cloud egress settings do not extend to self-hosted machines",
        "Conversation history is indefinite by default; deleting a transcript and artifacts does not immediately delete environment snapshots",
        "Snapshots expire after 30 days of inactivity; starting or resuming extends that window",
        "Cursor privacy documentation differs on whether all cloud runs enforce Privacy Mode; verify the actual starting state before sensitive use",
        "Vendor documentation only; no hands-on reliability test or security audit"
      ],
      "runtime": [
        "Managed cloud agents execute tools in dedicated Cursor-hosted VMs with repository clones, dependencies and configured secrets",
        "My Machines and Team Pools execute tool calls on user-managed hardware, but the agent loop still runs in Cursor cloud",
        "Managed-cloud network policies do not cover self-hosted machines; those use their own firewall and network",
        "Cloud agents auto-run terminal commands and do not use local Run Modes"
      ],
      "privacy_security": "With Privacy Mode on, Cursor says cloud-agent code, prompts and responses are not used for training. This is not zero storage: conversation history is indefinite by default and can be deleted with its artifacts; environment snapshots cannot be deleted on demand and expire after 30 days of inactivity, extended on start/resume; Enterprise teams can cap conversation retention with retention policies. Secrets and tokens remain until removed. The security overview says cloud agents run in Privacy Mode, while Secrets & Network still describes disabled-mode collection and warns that enabling it mid-run does not change that run. Treat this documentation mismatch as a reason to check the actual starting policy. Most models have zero-data-retention agreements, but the overview notes exceptions; this check did not verify each model. These are vendor statements, not an independent audit.",
      "pricing": {
        "amount": 8,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Free $0 (limited agent requests); Plus $8/month; Premium $30/month (marked most popular); Super $100/month; Ultra $200/month; Teams from $30/user/month; Enterprise custom",
        "usage_charges": "Plans include usage allowances that vary by tier; third-party models are billed separately by usage",
        "verified_at": "2026-10-28",
        "source_url": "https://cursor.com/pricing",
        "note": "Cursor's pricing page (read 22 October 2026) lists Free at $0, Plus at $8/month, Premium at $30/month (marked most popular), Super at $100/month, Ultra at $200/month, Teams from $30/user/month and custom Enterprise. This replaces the earlier Individual $20/month and Teams $40/user/month structure. Plan inclusions are usage-based; third-party models are billed separately by usage. Re-read 28 October 2026: feature table lists 'Cloud agents & automations' on Premium ($30/month) and above - Free and Plus ($8) are excluded. Page now also lists code hosting via Origin, AI code reviews via Bugbot, billing via X Subscription Corp and a usage pool shared with Grok; third-party models remain at API list price with spending caps."
      },
      "sources": [
        "https://cursor.com/docs/cloud-agent",
        "https://cursor.com/pricing",
        "https://cursor.com/docs/cloud-agent/security",
        "https://cursor.com/docs/cloud-agent/security-network",
        "https://cursor.com/docs/agent/security/run-modes",
        "https://cursor.com/docs/cloud-agent/choose-runtime",
        "https://cursor.com/blog/agent-sdk"
      ],
      "changelog": [
        {
          "date": "2026-10-13",
          "change": "Added eight source-linked feature and limit rows plus a narrow start checklist from current vendor documentation. No hands-on test or ranking."
        },
        {
          "date": "2026-10-08",
          "change": "Checked six Cursor pages. Added command auto-run, network and secret scope, transcript/snapshot deletion differences, self-hosting boundary and privacy-documentation mismatch. Removed Free filter tag because Hobby does not list cloud agents."
        },
        {
          "date": "2026-10-03",
          "change": "Added from a user-supplied shortlist after checking cursor.com/pricing and the cloud agents docs. Documentation research only."
        },
        {
          "date": "2026-10-22",
          "change": "Feature/access/privacy fields re-verified against cursor.com/docs/cloud-agent/security and security-network. Correction: VM snapshot retention is now a rolling 30 days of inactivity (was documented as 90); Enterprise retention policies can cap conversation history. Access inheritance, egress modes, secret scoping, auto-run behaviour and the Privacy Mode documentation mismatch re-confirmed."
        },
        {
          "date": "2026-10-28",
          "change": "Field-level pricing audit: cursor.com/pricing re-read - feature table shows cloud agents from Premium $30/month (Free/Plus excluded); availability and price_summary corrected, meta title set to from-$30; billing via X Subscription Corp with Grok usage pool noted. checked_label unchanged (not a full-record check)."
        },
        {
          "date": "2026-10-29",
          "change": "Vendor blog (20 October 2026) documents the Agent SDK: define agents in code and deploy to Cursor infrastructure with `agent-sdk deploy`; the framework behind managed agents such as Bugbot. Noted as an adjacent offering to the cloud-agents product; no field changes, checked_label unchanged."
        }
      ],
      "coverage_rows": [
        [
          "Repository access",
          "Admin and user scope",
          "Git provider app repository selection and triggering-user access both apply. Protected Git Scopes and a blocklist can narrow access.",
          "https://cursor.com/docs/cloud-agent/security"
        ],
        [
          "Command approval",
          "Auto-run in cloud",
          "Cloud agents run terminal commands without per-command approval. Local Run Modes are not cloud gates; review the draft pull request before merging.",
          "https://cursor.com/docs/cloud-agent/security"
        ],
        [
          "Network policy",
          "Configurable, not absolute",
          "Allow all, Default + allowlist and Allowlist only apply to managed cloud. Essential Cursor/SCM domains remain accessible even in allowlist-only mode.",
          "https://cursor.com/docs/cloud-agent/security-network"
        ],
        [
          "Secret types",
          "Different exposure",
          "Environment Variables are agent-visible. Runtime Secrets are redacted from outputs/commits but visible in Terminal; Build Secrets are build-only.",
          "https://cursor.com/docs/cloud-agent/security-network"
        ],
        [
          "Transcript deletion",
          "Separate from snapshots",
          "Conversation history is indefinite by default and deletable. VM snapshots cannot be deleted on demand and expire after 90 inactive days, reset on start/resume.",
          "https://cursor.com/docs/cloud-agent/security-network"
        ],
        [
          "Privacy Mode",
          "Documentation mismatch",
          "Security overview says cloud runs use Privacy Mode, while the configuration reference still describes collection when it is disabled and says enabling it mid-run does not change that run. Check the starting policy and model-specific exceptions.",
          "https://cursor.com/docs/cloud-agent/security-network"
        ],
        [
          "Self-hosted work",
          "Not a local agent loop",
          "My Machines and Enterprise Team Pools move tool execution to owner hardware, but the agent loop stays in Cursor cloud. Managed-cloud egress controls do not cover those workers.",
          "https://cursor.com/docs/cloud-agent/choose-runtime"
        ],
        [
          "Paid access",
          "Usage limits apply",
          "Cloud agents are listed on paid Individual and Teams plans, not Hobby. Included model usage has limits; on-demand usage is billed in arrears and tax is extra.",
          "https://cursor.com/pricing"
        ]
      ],
      "setup_steps": [
        "Use one non-sensitive repository and a task with acceptance tests.",
        "Restrict Git app scope, select the runtime and configure exact outbound domains.",
        "Check starting Privacy Mode, secret type, transcript/snapshot retention and usage budget.",
        "Review the draft pull request, tests and artifacts before a human merges it."
      ],
      "price_tags": [
        "free",
        "listed",
        "usage"
      ],
      "price_summary": "Free $0; Plus $8; Premium $30 (cheapest tier with cloud agents); Super $100; Ultra $200/month; Teams from $30/user (vendor pricing)",
      "meta_title": "Cursor cloud agents pricing (Oct 2026): from $30/month",
      "page_url": "https://agentcomparison.net/agents/cursor/"
    },
    {
      "id": "jules",
      "name": "Jules",
      "vendor": "Google",
      "kind": "product",
      "category": "coding agent (Google)",
      "group": "coding-agents",
      "summary": "Jules is Google's asynchronous GitHub coding agent. In its web workflow, you select a repository and branch, describe a task, approve a plan, then review the changes and publish a pull request. Code runs in a cloud VM. The API also supports automated workflows, but its approval defaults differ from the web app.",
      "one_liner": "Google's background coding agent that turns GitHub tasks into pull requests.",
      "best_fit": "GitHub developers with a working project setup who want bounded bug fixes, dependency maintenance, documentation or tests returned for review while they work on something else.",
      "poor_fit": "Teams needing a confirmed Workspace paid upgrade path, repositories with embedded secrets, or workflows that cannot tolerate an internet-connected cloud VM. Not a substitute for testing and reviewing generated changes.",
      "decision_note": "Best for scoped repository maintenance, tests and documentation that you can review as a pull request. Start with a non-sensitive repository, limit the GitHub app to selected repos, and provide a working setup script. Web plans can be reviewed before code changes; API sessions automatically approve plans unless requirePlanApproval is true. Automatic PR creation is a separate API setting, not automatic merging.",
      "availability": "Sign in with a Google account, then connect the Jules GitHub app to all or selected repositories. Google documents a no-cost plan. Paid Jules in Pro and Ultra benefits come through Google AI plans and currently require individual @gmail.com accounts; Workspace upgrade paths are not yet documented as available. Users must be 18 or older. Region eligibility was not verified in this check.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Web plan approval, cloud VM changes and reviewed pull request workflow",
          "source_url": "https://jules.google/docs/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Task and concurrency limits apply",
          "region_restrictions": "Not verified"
        },
        {
          "name": "Assign a task through a GitHub issue with a 'jules' label",
          "source_url": "https://jules.google/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Connected GitHub repositories",
          "region_restrictions": "Not verified"
        },
        {
          "name": "API sessions with explicit plan-approval and automatic-PR settings",
          "source_url": "https://developers.google.com/jules/api",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Connected sources required; default API plan approval is automatic",
          "region_restrictions": "Not verified"
        },
        {
          "name": "No-cost tier and Google AI Pro / Ultra task allowances",
          "source_url": "https://jules.google/docs/usage-limits/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Paid benefits currently for individual @gmail.com accounts; 18+",
          "region_restrictions": "Not verified"
        }
      ],
      "permissions": [
        "Connect the Google Labs Jules GitHub app to all or selected repositories. The FAQ explains changing this scope through GitHub Settings > Applications > Google Labs Jules > Configure. Refresh Jules after changing repository access.",
        "The getting-started web workflow shows a plan to review and approve before code changes. Review the diff and tests before publishing or merging the resulting pull request.",
        "API sessions automatically approve plans unless requirePlanApproval is true. automationMode=AUTO_CREATE_PR can create a pull request when changes are ready; the default does not automatically create one. These API controls do not authorize merging."
      ],
      "limitations": [
        "Requires a Google sign-in and GitHub repository access; select only the repositories needed",
        "Tasks are capped in a rolling 24-hour window: 15 / 100 / 300, with 3 / 15 / 60 concurrent tasks",
        "Paid upgrades currently documented for individual @gmail.com accounts, not Workspace accounts; age 18+",
        "API plans are automatically approved by default; set requirePlanApproval=true when approval is required",
        "Cloud VM has internet access; setup scripts, dependencies and non-code files need security review",
        "Vendor documentation only; AgentComparison has not tested task quality or reliability"
      ],
      "runtime": [
        "Internet-connected cloud VM that clones a selected GitHub repository",
        "Web app, GitHub issue label, or API sessions using a connected source",
        "Optional setup scripts and root AGENTS.md instructions"
      ],
      "privacy_security": "Re-verified 22 October 2026: Google's Jules FAQ states verbatim 'Jules does not train on private repository content... we do not use your private repositories to train models.' Each task runs in a fresh cloud VM that clones your repo with internet access; Google advises treating it like any shared compute surface - do not commit secrets to repos you connect, and review non-code files too. Repository access is manageable through the GitHub app connection. A linked page describes separate data use 'to improve Jules'; the training exclusion applies to private repo content. Vendor statements, not independently audited.",
      "pricing": {
        "amount": null,
        "currency": "USD",
        "billing_period": null,
        "allowances": "No-cost Jules: 15 tasks per rolling 24 hours, 3 concurrent. Jules in Pro: 100 tasks, 15 concurrent. Jules in Ultra: 300 tasks, 60 concurrent.",
        "usage_charges": "Task limits apply; the FAQ confirms a no-cost plan. Local Google AI subscription prices were not verified.",
        "verified_at": "2026-10-22",
        "source_url": "https://jules.google/docs/usage-limits/",
        "note": "Jules in Pro and Ultra are included in corresponding Google AI plans. No standalone Jules subscription price is listed in these docs; paid benefits currently require individual @gmail.com accounts. Exact local subscription price remains unknown."
      },
      "sources": [
        "https://jules.google/",
        "https://jules.google/docs/",
        "https://jules.google/docs/usage-limits/",
        "https://jules.google/docs/faq/",
        "https://developers.google.com/jules/api",
        "https://jules.google/docs/api/reference/types/"
      ],
      "changelog": [
        {
          "date": "2026-10-13",
          "change": "Added eight source-linked feature and limit rows plus a narrow start checklist from current vendor documentation. No hands-on test or ranking."
        },
        {
          "date": "2026-10-08",
          "change": "Checked six Google sources. Corrected Google sign-in versus GitHub connection, confirmed no-cost plan and paid account eligibility, and separated API auto-approval from web plan review. Added private-repository training scope and cloud VM precautions."
        },
        {
          "date": "2026-10-03",
          "change": "Added from a user-supplied shortlist after reading jules.google. Prices not stated on the page; recorded as unknown."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: no-training-on-private-repos claim re-verified verbatim in the current FAQ (jules.google/docs/faq/); fresh-VM-per-task and secrets guidance confirmed; note: jules.google.com/faq currently renders an unrelated 'Stitch Loop' page - the docs subdomain path is the live FAQ."
        }
      ],
      "coverage_rows": [
        [
          "Account and repository",
          "Two connections",
          "Sign in with Google, then connect the Jules GitHub app to selected repositories. Paid benefits currently require an individual @gmail.com account; users must be 18+.",
          "https://jules.google/docs/usage-limits/"
        ],
        [
          "Web plan review",
          "Documented workflow",
          "Getting started shows Give me a plan, review and approve before code changes. This is not proof that every integration uses that gate.",
          "https://jules.google/docs/"
        ],
        [
          "API approval",
          "Automatic by default",
          "API plans are automatically approved unless requirePlanApproval is true. Explicit approval must be configured, not assumed from the web workflow.",
          "https://developers.google.com/jules/api"
        ],
        [
          "Pull request automation",
          "Separate choice",
          "AUTO_CREATE_PR can create a PR when changes are ready. The default creates no PR automatically; neither setting is permission to merge.",
          "https://jules.google/docs/api/reference/types/"
        ],
        [
          "Environment setup",
          "Discrete commands",
          "Cloud VMs clone the repository and run setup. Long-lived dev servers/watch scripts are not supported in setup; root AGENTS.md supplies project context.",
          "https://jules.google/docs/faq/"
        ],
        [
          "Network and dependencies",
          "Security review needed",
          "VM has internet access. Review dependencies, shell commands and non-code files, and do not keep secrets in the repository.",
          "https://jules.google/docs/faq/"
        ],
        [
          "Training scope",
          "Private-repo statement",
          "FAQ says private repository content is not used for training. It does not establish all prompt, feedback, public-repository or retention terms.",
          "https://jules.google/docs/faq/"
        ],
        [
          "Task allowances",
          "Rolling limits",
          "No-cost/Pro/Ultra list 15/100/300 tasks per rolling 24 hours and 3/15/60 concurrent tasks. Paid local Google AI subscription prices remain unverified.",
          "https://jules.google/docs/usage-limits/"
        ]
      ],
      "setup_steps": [
        "Connect one non-sensitive GitHub repository using a Google account.",
        "Use a working discrete install/test setup and one clearly bounded task.",
        "Review the web plan; for API work explicitly set requirePlanApproval=true if required.",
        "Review the diff and test results before publishing or merging; check the task allowance separately."
      ],
      "price_tags": [
        "free"
      ],
      "price_summary": "No-cost plan: 15 tasks per rolling 24 hours; higher limits via Google AI Pro / Ultra, local subscription prices unverified",
      "page_url": "https://agentcomparison.net/agents/jules/"
    },
    {
      "id": "folk",
      "name": "Folk",
      "vendor": "Folk",
      "kind": "product",
      "category": "personal agent in iMessage, Telegram and WhatsApp",
      "group": "personal-assistants",
      "summary": "Folk is an assistant you text on iMessage, Telegram or WhatsApp. Its FAQ describes habit and goal check-ins, email search, web browsing, reminders, meeting notes, phone calls to your own number and background tasks. As of 22 October 2026 its pricing page says folk is free with every feature included, subject to a 100-message daily limit.",
      "one_liner": "A texting coach and assistant that tracks goals, email and reminders, now free with every feature included.",
      "best_fit": "People who want a texting assistant that nudges them on goals and habits and connects to common work apps.",
      "poor_fit": "Anyone who needs an independently audited security attestation; we did not find one in this check.",
      "decision_note": "Folk is now free (100 messages a day, every feature included) rather than $9.99 a month, which removes the cost barrier to trying it with one reminder or goal check-in and minimum account access. Its referral programme pays $15 cash per qualifying signup. The pricing-page FAQ says messages are never sold and never used to train AI models, with one-tap account wipe; third-party calling details remain inconsistent across official documents, so do not depend on them.",
      "availability": "iMessage and Telegram documented; WhatsApp is switching on country by country, per the FAQ. Check local signup availability.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Texts reminders, goal check-ins and habit streaks",
          "source_url": "https://www.folk.com/faq",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Connects Gmail, Google Calendar, Notion, Slack and more",
          "source_url": "https://www.folk.com/faq",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Calls your own number (10 minutes a week included)",
          "source_url": "https://www.folk.com/pricing",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Calls companies on your behalf with your approval (businesses only, discloses it is an AI, does not use plan minutes)",
          "source_url": "https://www.folk.com/docs/calls",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Joins iMessage or Telegram group chats and plans when tagged",
          "source_url": "https://www.folk.com/pricing",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Folk's FAQ lists one-tap connections to Gmail, Google Calendar, Drive, Docs, Sheets, Outlook, Teams, Notion, Slack, Linear, Todoist, Strava and bank accounts. These are vendor statements; review each connection you grant.",
        "The privacy policy documents optional Google read-only access and read-only Plaid bank data. Check every account permission; disconnect does not delete previously saved emails/files/notes."
      ],
      "limitations": [
        "Free background work is limited to 3 simultaneous tasks; one-off reminders do not count, per FAQ.",
        "Company-calling is documented with consent gating (folk only dials after you say yes, businesses only, numbers from the company's own website); reliability of those calls is untested. The calls documentation still lists a Pro minutes tier (120/month) that no longer appears on the pricing page.",
        "The pricing.md mirror conflicts with the live pricing page and FAQ on prices and phone allowances.",
        "Vendor documentation only; no reliability or independent security audit is claimed."
      ],
      "runtime": [
        "Messaging: iMessage, Telegram, WhatsApp",
        "Folk says it runs on its own private cloud computer (TechCrunch, 3 Oct 2026)"
      ],
      "privacy_security": "The privacy policy (updated 20 October 2026) says no AI-model training and no data sales. Model requests route through OpenRouter to providers including Anthropic, OpenAI, Google, Together AI, Fireworks AI, Z.ai and Sail Research on paid API terms that prohibit training; zero-data-retention routing covers eligible requests only, not every connected service. Observability vendors (Langfuse, Lemma, Raindrop) process message content for reliability, and OpenAI processes voice-call audio. Direct Google connections can be read-only; Plaid bank access is read-only and cannot move money. Folk runs on hosted agent servers, with a dedicated cloud sandbox for some accounts and temporary computers for some tasks. Disconnecting an app stops new access but saved content remains until deleted. Account deletion (settings, then privacy) revokes access and removes account data with external cleanup retried; encrypted backups expire on their retention schedule and limited records may be kept for payments, taxes, security or legal obligations. Optional iPhone connections can sync a screenshot library and match contacts via one-way hashes (opt out by email). These are vendor claims, not an independent audit.",
      "pricing": {
        "amount": 0,
        "currency": "USD",
        "billing_period": null,
        "allowances": "100 messages a day, reset at local midnight; every feature included: apps, memory, browsing, research, the builder, Gmail/Calendar/bank and 1,000+ app connections, 3 background tasks, 10 minutes of phone calls a week",
        "usage_charges": null,
        "verified_at": "2026-10-22",
        "source_url": "https://www.folk.com/pricing",
        "note": "Folk's pricing page (read 22 October 2026) now says 'folk is free': $0 a month, no card or subscription, 100 messages a day resetting at local midnight, with every feature included (apps, memory, browsing, research, the builder, 3 background tasks, 10 minutes of phone calls a week). The earlier Pro tier ($9.99/month, $99.99/year, verified 10 October 2026) is no longer published. A referral programme pays $15 cash when an invited friend signs up, sends 15 messages and passes checks.",
        "class": "free"
      },
      "sources": [
        "https://www.folk.com/faq",
        "https://www.folk.com/docs",
        "https://techcrunch.com/2026/10/03/all-the-ai-agents-that-can-live-in-your-text-messages/",
        "https://www.folk.com/privacy",
        "https://www.folk.com/docs/calls",
        "https://www.folk.com/pricing"
      ],
      "changelog": [
        {
          "date": "2026-10-10",
          "change": "Expanded source-linked feature evidence and narrow setup; clarified access, permissions, costs and unresolved limits. Documentation research only, not a product test."
        },
        {
          "date": "2026-10-03",
          "change": "Added from the TechCrunch roundup of 3 October 2026 (used as a lead only) after checking the vendor's own pages. Documentation research only."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: privacy policy re-read (updated 20 October 2026; OpenRouter provider routing, observability vendors, ZDR limits, iPhone screenshot/contact features, deletion flow). Company-calling and group-chat capabilities added from docs/calls and pricing FAQ; call limitation corrected (company calls are documented with consent gating)."
        }
      ],
      "coverage_rows": [
        [
          "Text channels",
          "Documented, rollout varies",
          "iMessage and Telegram are listed. The FAQ says WhatsApp is switching on country by country, so do not promise universal WhatsApp access.",
          "https://www.folk.com/faq"
        ],
        [
          "Background tasks",
          "Documented",
          "Vendor FAQ: up to 3 background tasks on Free, unlimited on Pro; one-off reminders do not count. No reliability test was run.",
          "https://www.folk.com/faq"
        ],
        [
          "App connections",
          "Documented",
          "Gmail, Calendar, Drive, Docs, Sheets, Outlook, Teams, Notion, Slack, Linear, Todoist, Strava and bank connections are listed, plus MCP tools. Review the permission for each account.",
          "https://www.folk.com/faq"
        ],
        [
          "Read-only access",
          "Documented",
          "Privacy policy says direct Google connections can be read-only. Plaid bank data is read-only balances/transactions; Folk says it cannot move money.",
          "https://www.folk.com/privacy"
        ],
        [
          "Phone calls",
          "Own-number calls documented; company calls conflict",
          "FAQ and privacy policy limit calls to the user's verified number. Calls documentation separately advertises calls to companies after approval. Third-party calling remains unverified due to this conflict.",
          "https://www.folk.com/docs/calls"
        ],
        [
          "Training and processing",
          "Vendor policy, not audit",
          "Policy says no AI model training on user data. AI providers and observability services still process message content; no-training does not mean no processing or human access.",
          "https://www.folk.com/privacy"
        ],
        [
          "Disconnect and deletion",
          "Documented with exceptions",
          "Disconnect stops new app access, but saved emails/files/notes remain until deleted. Account deletion removes live data; backups and required legal records can remain, with no exact backup deadline stated.",
          "https://www.folk.com/privacy"
        ],
        [
          "Price sources",
          "Live price verified; mirror conflict",
          "FAQ and live pricing page show Pro $9.99/month or $99.99/year. The pricing.md mirror differs on price and limits; use the current live checkout offer before paying, not the conflicting mirror.",
          "https://www.folk.com/pricing"
        ]
      ],
      "setup_steps": [
        "Start through the official site and a supported texting channel.",
        "Use one reminder or goal check-in before adding account access.",
        "Inspect each connected account's allowed actions; choose Google read-only if that fits the task.",
        "Review active background tasks and memory in settings. Disconnecting an app is not the same as deleting saved content."
      ],
      "price_tags": [
        "free"
      ],
      "price_summary": "Free ($0), 100 messages a day, every feature included; $15 referral programme (vendor pricing page)",
      "page_url": "https://agentcomparison.net/agents/folk/"
    },
    {
      "id": "caddy",
      "name": "Caddy",
      "vendor": "Caddy",
      "kind": "product",
      "category": "personal assistant in iMessage and RCS",
      "group": "personal-assistants",
      "summary": "Caddy is an AI assistant you text. On iPhone it uses iMessage, on Android RCS and SMS. Its site describes adding events to calendars, setting reminders, tracking follow-ups and doing research, and says you can use voice memos or add it to group chats.",
      "one_liner": "A text-message assistant for calendar, reminders and follow-ups, in public beta.",
      "best_fit": "People who want to handle calendar, reminders and follow-ups from a text thread without another app.",
      "poor_fit": "Anyone who needs a published price or detailed security documentation before connecting accounts.",
      "decision_note": "Caddy's site does not publish a price on the pages we read, and TechCrunch says it has been in public beta since April 2026. Its privacy answer is brief, so read the full policy before connecting accounts.",
      "availability": "Texting on iPhone and Android per the vendor FAQ. Pricing and region details not found.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Calendar events, reminders and follow-ups by text",
          "source_url": "https://caddy.app/",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Works in iMessage, RCS and SMS, with voice memos and group chats",
          "source_url": "https://caddy.app/",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Adds events to your calendar, sets reminders and keeps an eye on calendars, chats and tasks to flag what needs attention.",
        "Proactive: 'spots things before you have to. It can make plans, set up tasks, or just do it if you can't' - the site does not describe an approval step.",
        "Can act in connected apps: examples show booking a restaurant ('Want me to book it?') and filing notes into a project brief.",
        "Joins group chats when dropped in."
      ],
      "limitations": [
        "No price found on the pages we read",
        "Public beta per TechCrunch (3 Oct 2026)",
        "Privacy statement on the site is short and says data personalises and improves the experience",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Messaging: iMessage, RCS, SMS, group chats"
      ],
      "privacy_security": "Caddy's only published privacy statement is the homepage FAQ: 'Caddy is built on a simple principle: your messages and data should stay private. We use your data only to personalize and improve your experience.' No standalone privacy policy, retention schedule, deletion route or training-use statement was found on caddy.app as of 23 October 2026. Messages necessarily pass through iMessage/RCS and Caddy's messaging provider (the FAQ notes provider issues can interrupt service).",
      "pricing": {
        "amount": null,
        "currency": "USD",
        "billing_period": null,
        "allowances": null,
        "usage_charges": "Not stated",
        "verified_at": "2026-10-21",
        "source_url": "https://caddy.app/",
        "note": "No price or plan list was found on caddy.app. TechCrunch only reports a public beta since April 2026.",
        "class": "see_vendor"
      },
      "sources": [
        "https://caddy.app/",
        "https://techcrunch.com/2026/10/03/all-the-ai-agents-that-can-live-in-your-text-messages/"
      ],
      "changelog": [
        {
          "date": "2026-10-03",
          "change": "Added from the TechCrunch roundup of 3 October 2026 (used as a lead only) after checking the vendor's own pages. Documentation research only."
        },
        {
          "date": "2026-10-21",
          "change": "Re-checked vendor site: no pricing published, claims unchanged."
        }
      ],
      "coverage_rows": [
        [
          "Channel",
          "iMessage / RCS",
          "On iPhone, Caddy texts with blue bubbles through iMessage; on Android it texts through RCS. No app to download; voice memos and group chats supported.",
          "https://caddy.app/"
        ],
        [
          "Availability",
          "Public beta",
          "FAQ: Caddy is currently in public beta; 'as with any AI, Caddy can make mistakes, so please verify information and actions.'",
          "https://caddy.app/"
        ],
        [
          "Privacy",
          "FAQ statement only",
          "FAQ: 'your messages and data should stay private. We use your data only to personalize and improve your experience.' No standalone privacy policy was found on caddy.app.",
          "https://caddy.app/"
        ],
        [
          "Pricing",
          "Not published",
          "No price or plan list appears on caddy.app (checked 23 October 2026); the beta is usable without a published price.",
          "https://caddy.app/"
        ]
      ],
      "setup_steps": [
        "Text Caddy from your own number - no app to download.",
        "On iPhone it uses iMessage; on Android, RCS.",
        "Send texts or voice memos; you can drop Caddy into group chats.",
        "Support: contact@trycaddy.ai."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Price not published on the pages we read",
      "page_url": "https://agentcomparison.net/agents/caddy/"
    },
    {
      "id": "fambot",
      "name": "Fambot",
      "vendor": "Fambot",
      "kind": "product",
      "category": "family chief-of-staff assistant by text",
      "group": "personal-assistants",
      "summary": "Fambot turns school emails, newsletters and calendar updates into a daily plan. Its site says it texts a plan each evening and that you can reply with questions or requests. It connects to Gmail and Google Calendar.",
      "one_liner": "Turns school emails and calendars into a nightly plan by text, for parents.",
      "best_fit": "Parents who get school and activity information by email and want it as a short daily plan.",
      "poor_fit": "Anyone who needs Outlook or Apple Calendar support now; TechCrunch says it is planned.",
      "decision_note": "It is free during beta. Fambot says it will later offer a free and a paid tier but does not state prices, so we list price as unknown. TechCrunch (3 Oct 2026) says it expects to charge roughly the cost of a Netflix subscription; that is not a vendor price.",
      "availability": "iOS, Android, web and SMS per the vendor site and TechCrunch.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Nightly plan by text for the next day",
          "source_url": "https://fambot.com/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Finds dates in family emails and offers them for your calendar",
          "source_url": "https://fambot.com/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Works with Gmail and Google Calendar: connects once, then reads school emails, newsletters and calendar updates automatically - no forwarding, tagging or manual sorting.",
        "Outlook, iCloud and other providers are not yet supported; the site offers an early-access list for them.",
        "Vendor states it never sends emails and never edits the calendar without permission.",
        "No public statement on data retention, deletion or model-training use as of 23 October 2026."
      ],
      "limitations": [
        "Beta; paid plan and prices not yet announced",
        "Connects to Gmail and Google Calendar; others planned (TechCrunch)",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Messaging: SMS, plus iOS, Android and web apps"
      ],
      "privacy_security": "Fambot's site states: 'We never send emails, never edit your calendar without permission, and never sell or share your data.' It reads connected Gmail and Google Calendar content (school emails, newsletters, calendar updates) to produce a daily family plan. No separate training-use or retention statement was found on the public site as of 23 October 2026.",
      "pricing": {
        "amount": null,
        "currency": "USD",
        "billing_period": null,
        "allowances": null,
        "usage_charges": "Free while in beta; later a free tier and a paid tier (vendor FAQ)",
        "verified_at": "2026-10-22",
        "source_url": "https://fambot.com/",
        "note": "Fambot's FAQ says no credit card is required and it is free in beta. Later prices are not published.",
        "class": "free"
      },
      "sources": [
        "https://fambot.com/",
        "https://techcrunch.com/2026/09/01/fambot-introduces-an-ai-chief-of-staff-for-families/",
        "https://techcrunch.com/2026/10/03/all-the-ai-agents-that-can-live-in-your-text-messages/"
      ],
      "changelog": [
        {
          "date": "2026-10-23",
          "change": "Field-level audit from live site: free in beta (no card), free+paid tiers planned; Gmail/Google Calendar connect-once model (Outlook/iCloud waitlisted); vendor states never sends emails, never edits calendar without permission, never sells/shares data; no public training/retention statement found (recorded as unknown, not assumed). Coverage rows, permissions, setup steps filled. No hands-on testing."
        },
        {
          "date": "2026-10-22",
          "change": "Re-verified live site FAQ: no credit card required, free while in beta, free and paid tiers promised later (no prices published). Unchanged. No hands-on testing."
        },
        {
          "date": "2026-10-03",
          "change": "Added from the TechCrunch roundup of 3 October 2026 (used as a lead only) after checking the vendor's own pages. Documentation research only."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Free in beta",
          "Site FAQ: free while in beta, no credit card required; a free tier and a paid tier are planned later, with existing users keeping access.",
          "https://fambot.com/"
        ],
        [
          "Account access",
          "Gmail + Google Calendar",
          "Connects once to Gmail and Google Calendar; Outlook/iCloud waitlisted. Reads content to extract deadlines, early-release days and conference slots.",
          "https://fambot.com/"
        ],
        [
          "Action limits",
          "No sends, permission-gated edits",
          "Vendor states it never sends emails and never edits the calendar without permission. Vendor statement, not tested.",
          "https://fambot.com/"
        ],
        [
          "Data use",
          "No sale or sharing",
          "Vendor states it never sells or shares user data. No training/retention statement published.",
          "https://fambot.com/"
        ]
      ],
      "setup_steps": [
        "Connect one Gmail/Google Calendar account and check which school newsletters and calendars it starts reading.",
        "Review the first daily plan for over- or under-inclusion before relying on it for deadlines.",
        "Note that Outlook and iCloud accounts are not yet supported - join the early-access list only if you need them."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Free during beta; later prices not published (vendor FAQ)",
      "page_url": "https://agentcomparison.net/agents/fambot/"
    },
    {
      "id": "iris",
      "name": "Iris",
      "vendor": "Iris AI",
      "kind": "product",
      "category": "personal agent via iMessage, SMS and WhatsApp",
      "group": "personal-assistants",
      "summary": "Iris is an assistant you text on iMessage or SMS, with Android and browser access. Its agent mode can use email, calendar and browser tools, run scheduled check-ins and text results back. Plans are priced by monthly credits.",
      "one_liner": "A texting assistant for email, calendar, research and errands, with paid credit plans.",
      "best_fit": "People who want a texting assistant with scheduled tasks and several named agents.",
      "poor_fit": "Anyone who needs to keep content away from third-party AI model providers.",
      "decision_note": "Iris says 'free to start', but its pricing page lists paid plans with monthly credits. Content you send is processed by third-party model providers per its privacy policy. TechCrunch says Iris runs on Hermes agent; we did not verify that.",
      "availability": "iMessage, SMS, WhatsApp, Android app and browser per the vendor pages.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Texting assistant on iMessage and SMS",
          "source_url": "https://irisai.app/imessage",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Agent mode with email, calendar and web search tools (Premium)",
          "source_url": "https://irisai.app/pricing",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Premium and above",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Scheduled and proactive tasks",
          "source_url": "https://agents.irisai.app/",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Agent Mode (Premium and up) uses email, calendar and web-search tools, plus proactive outreach and scheduled tasks.",
        "Premium allows up to 5 projects and 10 agents; Pro is unlimited with team collaboration and shared agents.",
        "Pro adds model choice (Gemini, Claude, OpenAI and others) and priority response times.",
        "Pricing page claims 'private & encrypted conversations' (vendor-stated)."
      ],
      "limitations": [
        "Plans are credit-based: 200, 500 or 2,000 credits a month",
        "Conversation content generally retained up to 365 days (privacy policy)",
        "TechCrunch says Hermes agent powers Iris; not verified by us",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Messaging: iMessage, SMS, WhatsApp, Android app, web dashboard"
      ],
      "privacy_security": "Iris publishes a privacy policy: content is shared with AI providers only as needed to fulfil requests, and those providers may not use it to train their models except as permitted by Iris's agreements or law. Conversation content is generally retained up to 365 days, then automatically deleted; diagnostic and abuse-prevention logs are kept for a shorter period; closing your account or requesting deletion triggers deletion or de-identification, subject to legal retention and backup cycles. The policy grants access, correction, deletion and copy rights and states personal information is not sold or shared under CCPA/CPRA. Messaging via SMS/MMS/iMessage means your phone number and message content are processed. The pricing page additionally claims 'private & encrypted conversations' without detail. Checked 23 October 2026.",
      "pricing": {
        "amount": 11.99,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Starter 200 credits a month; Premium 500 credits a month; Pro 2,000 credits a month with rollover",
        "usage_charges": "Extra credits can be bought",
        "verified_at": "2026-10-23",
        "source_url": "https://irisai.app/pricing",
        "note": "irisai.app/pricing lists Starter at $11.99 a month, Premium at $23.99 a month ($19.19 on the annual view) and Pro at $99 a month. The site also says 'Get Started Free' without defining a free tier.",
        "class": "amount"
      },
      "sources": [
        "https://irisai.app/pricing",
        "https://irisai.app/privacy",
        "https://agents.irisai.app/",
        "https://techcrunch.com/2026/10/03/all-the-ai-agents-that-can-live-in-your-text-messages/",
        "https://irisai.app/imessage"
      ],
      "changelog": [
        {
          "date": "2026-10-03",
          "change": "Added from the TechCrunch roundup of 3 October 2026 (used as a lead only) after checking the vendor's own pages. Documentation research only."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "From $11.99/mo",
          "Starter $11.99/mo (200 credits); Premium $23.99/mo or $19.19/mo billed annually (500 credits, Agent Mode, email/calendar/web tools); Pro $99/mo (2,000 credits with rollover, model choice, unlimited agents).",
          "https://irisai.app/pricing"
        ],
        [
          "Channel",
          "iMessage / SMS",
          "Starter chats by text message with no app needed; agent dashboard at agents.irisai.app on Premium and above.",
          "https://irisai.app/pricing"
        ],
        [
          "Training use",
          "Prohibited by policy",
          "Privacy policy: providers may not use your content to train their models except as permitted by Iris's agreements or applicable law.",
          "https://irisai.app/privacy"
        ],
        [
          "Retention",
          "Up to 365 days",
          "Conversation content is generally retained up to 365 days, then automatically deleted; account data deleted or de-identified on closure or request.",
          "https://irisai.app/privacy"
        ],
        [
          "Privacy rights",
          "CCPA/CPRA + access rights",
          "Policy grants access, correction, deletion and copy rights; states it does not sell or share personal information under CCPA/CPRA.",
          "https://irisai.app/privacy"
        ]
      ],
      "setup_steps": [
        "Text Iris from your phone to start - Starter needs no app.",
        "Pick a plan: Starter $11.99/mo for chat; Premium ($23.99, or $19.19 billed annually) adds Agent Mode; Pro $99/mo for teams.",
        "Manage agents, projects and schedules from the dashboard at agents.irisai.app (Premium and above).",
        "Buy extra credits anytime; Pro credits roll over."
      ],
      "price_tags": [
        "listed",
        "usage"
      ],
      "price_summary": "Starter $11.99/month, Premium $23.99, Pro $99; credits per month (vendor pricing)",
      "page_url": "https://agentcomparison.net/agents/iris/"
    },
    {
      "id": "martin",
      "name": "Martin",
      "vendor": "Martin (trymartin.com)",
      "kind": "product",
      "category": "personal assistant by SMS, phone, WhatsApp and Slack",
      "group": "personal-assistants",
      "summary": "Martin is a personal assistant reachable by phone, SMS, WhatsApp, email and Slack. Its docs list calendar management, texting and calling others for bookings, reminders and web search. It also has an iOS app.",
      "one_liner": "An assistant you reach by call, text, WhatsApp, email or Slack for calendar, email and tasks.",
      "best_fit": "People who want one assistant across phone, text and chat tools.",
      "poor_fit": "Anyone who does not want an assistant contacting third parties for them.",
      "decision_note": "Martin acts on your behalf by texting or calling other people, so decide what it may send before connecting it. The pricing page lists Basic $21 a month billed yearly and Pro $30 a month billed yearly, each with a 7-day trial.",
      "availability": "Phone, SMS, WhatsApp, email, Slack and an iOS app per the vendor docs.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Reachable by phone, SMS, WhatsApp, email and Slack",
          "source_url": "https://docs.trymartin.com/introduction",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Texts and calls others for bookings and scheduling",
          "source_url": "https://docs.trymartin.com/introduction",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Calendar management and reminders",
          "source_url": "https://docs.trymartin.com/introduction",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Acts across phone, SMS, WhatsApp, email and Slack; can text and call other people for bookings and reminders.",
        "Proactive actions (limited on Basic, unlimited on Pro) and custom tasks; Basic caps at 2 tasks at a time.",
        "Uses Google Workspace APIs to execute tasks on behalf of users."
      ],
      "limitations": [
        "Basic plan: basic AI models, limited proactive actions, 2 tasks at a time",
        "Monthly billing costs more than the annual rate",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Phone call, SMS, WhatsApp, email, Slack, iOS app"
      ],
      "privacy_security": "Martin's privacy notice (info.trymartin.com/privacy) is a Termly-style policy. It includes a Google Limited Use disclosure: Google Workspace API user data is not used to develop, improve or train AI/ML models, and is not transferred to third-party AI tools for generalised model training. Retention is 'as long as necessary' for the stated purposes unless law requires longer; users can request access, correction, a copy or deletion. The notice discloses advertising use via retention.com with an opt-out link. No product-side data-use page or training statement beyond the Google clause was found as of 23 October 2026.",
      "pricing": {
        "amount": 21,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Basic: basic models, 2 tasks at a time. Pro: advanced models, unlimited proactive actions",
        "usage_charges": "Basic $21 a month billed yearly or $35 monthly; Pro $30 billed yearly or $49 monthly; 7-day trial",
        "verified_at": "2026-10-23",
        "source_url": "https://www.trymartin.com/pricing",
        "note": "trymartin.com/pricing (checked 23 October 2026): Personal plans - Basic $21/mo billed yearly or $35/mo monthly; Pro $30/mo billed yearly or $49/mo monthly. 7-day free trial; Black Friday banner offers 30% off the first year. Basic: basic models, limited proactive actions, 2 tasks at a time; Pro: most advanced models, unlimited proactive actions and custom tasks. A Business tab exists.",
        "class": "amount"
      },
      "sources": [
        "https://www.trymartin.com/pricing",
        "https://docs.trymartin.com/introduction",
        "https://info.trymartin.com/privacy",
        "https://techcrunch.com/2026/10/03/all-the-ai-agents-that-can-live-in-your-text-messages/"
      ],
      "changelog": [
        {
          "date": "2026-10-03",
          "change": "Added from the TechCrunch roundup of 3 October 2026 (used as a lead only) after checking the vendor's own pages. Documentation research only."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "From $21/mo",
          "Personal plans: Basic $21/mo billed yearly ($35 monthly); Pro $30/mo billed yearly ($49 monthly). 7-day free trial, cancel anytime; a Black Friday banner offers 30% off the first year.",
          "https://www.trymartin.com/pricing"
        ],
        [
          "Plan limits",
          "2 tasks vs unlimited",
          "Basic: basic AI models, limited proactive actions, 2 tasks at a time. Pro: the most advanced models, unlimited proactive actions and unlimited custom tasks.",
          "https://www.trymartin.com/pricing"
        ],
        [
          "Channels",
          "Phone/SMS/WhatsApp/email/Slack",
          "Reachable by phone call, SMS, WhatsApp, email and Slack, plus an iOS app.",
          "https://docs.trymartin.com/introduction"
        ],
        [
          "Training use",
          "Google data excluded",
          "Google Limited Use disclosure: Google Workspace user data is not used to develop, improve or train AI/ML models and is not transferred to third-party AI tools for generalised model training.",
          "https://info.trymartin.com/privacy"
        ],
        [
          "Retention",
          "As long as necessary",
          "Keeps personal information as long as necessary for the notice's purposes unless law requires longer; users may request access, correction, copy or deletion.",
          "https://info.trymartin.com/privacy"
        ]
      ],
      "setup_steps": [
        "Start the 7-day free trial; Basic $21/mo billed yearly ($35 monthly) or Pro $30/mo billed yearly ($49 monthly).",
        "Reach Martin on the channel you prefer: phone, SMS, WhatsApp, email or Slack; the iOS app is optional.",
        "Connect Google Workspace only after reading the Limited Use clause - Google data is excluded from model training."
      ],
      "price_tags": [
        "listed"
      ],
      "price_summary": "Basic $21/month billed yearly ($35 monthly); Pro $30 yearly ($49 monthly); 7-day trial (vendor pricing)",
      "page_url": "https://agentcomparison.net/agents/martin/"
    },
    {
      "id": "miso",
      "name": "Miso",
      "vendor": "Miso",
      "kind": "product",
      "category": "iMessage travel concierge",
      "group": "errands-travel-food",
      "summary": "Miso describes itself as a premium iMessage-based travel concierge. It says you can book flights over text with a dedicated team that knows your preferences and loyalty points, with seat selection and changes handled 24/7.",
      "one_liner": "A premium iMessage travel concierge for flights, with a human team behind it.",
      "best_fit": "Frequent flyers who want to hand flight booking and changes to a concierge by text.",
      "poor_fit": "Anyone who needs published pricing or wants a fully automated tool.",
      "decision_note": "This is a concierge service with a human team, not only software. We did not find a price on miso.com, so it is listed as unknown. 'Approved by Apple' is Miso's own claim.",
      "availability": "iMessage; an iOS app called Miso Intelligence also exists on the App Store. Region details not found.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Flight booking over iMessage with a dedicated team",
          "source_url": "https://miso.com/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Applies loyalty status and points; handles changes and cancellations",
          "source_url": "https://miso.com/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "iMessage-based: booking and changes happen over text with what the vendor describes as a dedicated team (human-in-the-loop concierge, not only software).",
        "The vendor says it applies your loyalty status and points and handles seat selection, changes and cancellations - implying access to airline accounts or booking identities, though no permission model is published.",
        "Distribution is via an iOS app ('Miso Intelligence' on the App Store) with iMessage as the conversation channel."
      ],
      "limitations": [
        "No price found on the page we read",
        "'World's first travel agent approved by Apple' is the vendor's own claim",
        "Human team involved per the vendor",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Messaging: iMessage"
      ],
      "privacy_security": "Miso's public site carries no privacy or data-use statement as of 23 October 2026. The service books real flights and applies loyalty status and points, which implies handling travel identity, payment and loyalty-account details, but no public statement covers how these are stored or used. Treat data handling as unpublished.",
      "pricing": {
        "amount": null,
        "currency": "USD",
        "billing_period": null,
        "allowances": null,
        "usage_charges": "Not stated",
        "verified_at": "2026-10-22",
        "source_url": "https://miso.com/",
        "note": "miso.com shows no price on the page we read."
      },
      "sources": [
        "https://miso.com/",
        "https://techcrunch.com/2026/10/03/all-the-ai-agents-that-can-live-in-your-text-messages/"
      ],
      "changelog": [
        {
          "date": "2026-10-23",
          "change": "Field-level audit from live site: iMessage-based premium travel concierge with dedicated team; full-trip handling (loyalty, seats, changes, cancellations, 24/7, direct airline integrations) as vendor claims; 'world's first travel agent approved by Apple' kept strictly vendor-attributed (no independent confirmation); pricing and privacy unpublished (recorded as unknown, not assumed). Coverage rows, permissions, setup steps filled. No hands-on testing."
        },
        {
          "date": "2026-10-22",
          "change": "Re-verified miso.com: still describes a premium iMessage travel concierge with dedicated team, loyalty/points handling, 24/7 changes, direct airline integrations and an App Store app ('world's first travel agent approved by Apple' is the vendor's own claim). No price published. No hands-on testing."
        },
        {
          "date": "2026-10-03",
          "change": "Added from the TechCrunch roundup of 3 October 2026 (used as a lead only) after checking the vendor's own pages. Documentation research only."
        }
      ],
      "coverage_rows": [
        [
          "Service model",
          "iMessage travel concierge",
          "Premium iMessage-based travel concierge: book flights over text with a dedicated team; handles full trip - itinerary, loyalty points/status, seat selection, changes and cancellations, 24/7, with direct airline integrations (vendor claims).",
          "https://miso.com/"
        ],
        [
          "Apple approval claim",
          "Vendor-attributed only",
          "Miso markets itself as 'the world's first travel agent approved by Apple'. This claim appears only in Miso's own materials; AgentComparison has not found independent Apple confirmation and presents it strictly as a vendor claim.",
          "https://miso.com/"
        ],
        [
          "Pricing",
          "Not published",
          "No price, subscription or fee information is published on the site; the product is described as 'premium' and distributed via the App Store.",
          "https://miso.com/"
        ],
        [
          "Privacy",
          "Not published",
          "No privacy or data-use statement on the public site; data handling for bookings, payments and loyalty accounts is unpublished.",
          "https://miso.com/"
        ]
      ],
      "setup_steps": [
        "Join from the site or install Miso Intelligence from the App Store and start the conversation in iMessage.",
        "Before sharing loyalty numbers or payment details, note that Miso publishes no privacy or data-handling statement - ask the team how your details are stored.",
        "Start with a low-stakes itinerary request and confirm how changes and cancellations are authorised before relying on it for time-critical travel."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Price not published on the page we read",
      "page_url": "https://agentcomparison.net/agents/miso/"
    },
    {
      "id": "ohai",
      "name": "Ohai",
      "vendor": "Ohai",
      "kind": "product",
      "category": "family household assistant by text",
      "group": "personal-assistants",
      "summary": "Ohai describes a text-based household assistant: you text, forward an email or speak a request and it turns it into plans, reminders and schedules. Its site lists calendar scanning, meal planning and lists, with human assistants stepping in for complicated items.",
      "one_liner": "A text-based household assistant for family calendars, reminders and lists, with a free version.",
      "best_fit": "Parents who want a household manager they can text, with optional human help.",
      "poor_fit": "Anyone who does not want a human reviewing complicated documents such as school schedules; the vendor says human assistants can step in.",
      "decision_note": "Ohai's own page says there is a free version and Premium plans start at $9.99 a month. Family-size pricing beyond that was not shown on the page we read. TechCrunch (3 Oct 2026) says plans scale with family size.",
      "availability": "iOS app and text per the vendor site. Region details not checked.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Turns texts, forwarded emails and voice requests into plans and reminders",
          "source_url": "https://www.ohai.ai/",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Calendar scan and shared household calendars (Premium)",
          "source_url": "https://www.ohai.ai/",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Premium",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Turns texts, forwarded emails and voice requests into plans, reminders and schedules.",
        "Human assistants can review complex items (school schedules, documents) - a manual-processing step on user content.",
        "Premium adds shared household calendars and family task coordination.",
        "If directed, Ohai may contact third parties to help execute a request."
      ],
      "limitations": [
        "Premium features are on paid plans from $9.99 a month (vendor)",
        "Human assistants may review complicated items (vendor)",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Messaging: text, email forwarding, voice, plus an app"
      ],
      "privacy_security": "Ohai's privacy policy (ohai.ai/privacy-policy/) describes a hybrid service: AI plus trained 'Human Assistants' who may review complicated user items. It uses collected information 'including training our algorithms to allow us to execute on your requests more efficiently' - the FAQ simultaneously says family information is used only to power features and is never sold or shared for advertising. Third-party service providers (e.g. payment processors) operate under the policy, and recommendations may carry retail affiliate identifiers, though personal data is not shared with retailers for affiliate purposes. Deletion route and retention schedule were not located in the extracted text as of 23 October 2026.",
      "pricing": {
        "amount": 9.99,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Free version with core features such as reminders, meal planning and lists; Premium unlocks shared household calendars and family task coordination",
        "usage_charges": "Premium plans start at $9.99 a month; higher family-size prices not shown on the page we read",
        "verified_at": "2026-10-23",
        "source_url": "https://www.ohai.ai/",
        "note": "ohai.ai FAQ (checked 23 October 2026): free version with core features (reminders, meal planning, lists); Premium plans from $9.99/month adding shared household calendars, family task coordination and household reminders.",
        "class": "amount"
      },
      "sources": [
        "https://www.ohai.ai/",
        "https://www.ohai.ai/privacy-policy/",
        "https://techcrunch.com/2026/10/03/all-the-ai-agents-that-can-live-in-your-text-messages/"
      ],
      "changelog": [
        {
          "date": "2026-10-03",
          "change": "Added from the TechCrunch roundup of 3 October 2026 (used as a lead only) after checking the vendor's own pages. Documentation research only."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Free + Premium from $9.99/mo",
          "Free version with core features (reminders, meal planning, lists). Premium plans from $9.99/mo add shared household calendars, family task coordination and household reminders.",
          "https://www.ohai.ai/"
        ],
        [
          "Human review",
          "Humans in the loop",
          "Ohai combines AI with trained human assistants who 'step in to review and organize' complicated items such as detailed school schedules or documents - user content may be seen by staff.",
          "https://www.ohai.ai/"
        ],
        [
          "Training use",
          "Trains own algorithms",
          "Privacy policy: information is used 'including training our algorithms to allow us to execute on your requests more efficiently'. FAQ says data is never sold or shared for advertising.",
          "https://www.ohai.ai/privacy-policy/"
        ],
        [
          "Affiliate links",
          "Present, no data shared",
          "Recommendations may link to third-party retailers with affiliate identifiers; the policy says personal data is not shared with retailers for affiliate purposes.",
          "https://www.ohai.ai/privacy-policy/"
        ]
      ],
      "setup_steps": [
        "Start free: text, forward an email, or speak a request.",
        "Premium from $9.99/mo adds shared household calendars and family task coordination.",
        "Know the human-in-the-loop step: complex items can be reviewed by Ohai staff."
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Free version; Premium from $9.99/month (vendor site)",
      "page_url": "https://agentcomparison.net/agents/ohai/"
    },
    {
      "id": "ollie",
      "name": "Ollie",
      "vendor": "Ollie",
      "kind": "product",
      "category": "family assistant by iMessage and SMS",
      "group": "personal-assistants",
      "summary": "Ollie is a text-first family assistant that watches your calendar and inbox, answers questions such as what is on this week and tells you what matters. You text a number over iMessage or SMS and connect Google or Outlook through standard OAuth consent.",
      "one_liner": "A text-first family assistant for calendar, inbox and school messages, billed by messages.",
      "best_fit": "Families who want a calendar-and-inbox assistant they text.",
      "poor_fit": "Anyone who needs a published SOC 2 report from the vendor's own site before connecting an inbox.",
      "decision_note": "Ollie bills by messages: a free plan with 50 messages a month, then $25 a month for 150 and $100 a month for 1,000, plus a small daily free allowance. TechCrunch says Ollie is SOC 2 compliant; we did not find that on the vendor pages we read, so it is not stated here as fact.",
      "availability": "iMessage and SMS per the vendor FAQ.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Watches calendar and inbox and texts what matters",
          "source_url": "https://ollie.ai/faq/",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Works over iMessage or SMS with Google or Outlook connections",
          "source_url": "https://ollie.ai/faq/",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Watches your calendar and inbox (Google or Outlook), answers questions and flags what matters.",
        "One subscription covers the whole family group chat.",
        "Upgrade, downgrade or cancel by texting Ollie - no portal needed."
      ],
      "limitations": [
        "Billing is by Ollie replies; plans cap messages per month",
        "SOC 2 claim comes from TechCrunch only; not verified by us",
        "Messages are sent to Anthropic, Google and OpenAI models via commercial APIs (vendor FAQ)",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Messaging: iMessage, SMS"
      ],
      "privacy_security": "Ollie's privacy policy states it does not use email or calendar content to train AI models, and that third-party model providers (models come from Anthropic, Google and OpenAI via commercial APIs per the FAQ) are contractually barred from training their public models on user content; de-identified or aggregated messages may be used. Connected-account data is retained until no longer necessary or until you disconnect; texting Ollie a deletion request deletes data from core services immediately and fully within 30 days, excluding backups purged on their cycle. CCPA rights apply with no sale or share for cross-context behavioral advertising. The site footer claims SOC 2 Type 1 compliance (trust.ollie.ai); legal entity Confabulation Corporation. Checked 23 October 2026.",
      "pricing": {
        "amount": 25,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Free: 50 messages a month plus 6 extra a day. $25: 150 a month plus 10 a day. $100: 1,000 a month plus 15 a day",
        "usage_charges": "Top-ups are sold when you run out (vendor pricing page)",
        "verified_at": "2026-10-23",
        "source_url": "https://ollie.ai/pricing/",
        "note": "ollie.ai/pricing (checked 23 October 2026): Free $0 (50 messages/mo + 6 free daily messages); Everyday $25/mo (150/mo + 10 daily); Always-On $100/mo (1,000/mo + 15 daily). One subscription covers the family group chat. A message = one Ollie reply. Upgrade/downgrade/cancel by text.",
        "class": "amount"
      },
      "sources": [
        "https://ollie.ai/pricing/",
        "https://ollie.ai/privacy-policy/",
        "https://ollie.ai/faq/",
        "https://techcrunch.com/2026/10/03/all-the-ai-agents-that-can-live-in-your-text-messages/"
      ],
      "changelog": [
        {
          "date": "2026-10-03",
          "change": "Added from the TechCrunch roundup of 3 October 2026 (used as a lead only) after checking the vendor's own pages. Documentation research only."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Free / $25 / $100",
          "Free $0 (50 messages/mo + 6 free daily); Everyday $25/mo (150/mo + 10 daily); Always-On $100/mo (1,000/mo + 15 daily). One subscription covers the whole family group chat. A 'message' is one Ollie reply.",
          "https://ollie.ai/pricing/"
        ],
        [
          "Channel",
          "iMessage / SMS",
          "Text a number over iMessage or SMS; connect Google or Outlook email and calendar.",
          "https://ollie.ai/pricing/"
        ],
        [
          "Training use",
          "Not used",
          "Policy: 'we do not use the content of your email or calendar to train AI models'; third-party model providers are contractually barred from training public models on user content. De-identified/aggregated messages may be used.",
          "https://ollie.ai/privacy-policy/"
        ],
        [
          "Deletion",
          "Text to delete",
          "Text Ollie to request deletion; data is deleted immediately from core services and removed within 30 days of a deletion request, excluding backups which purge on their cycle.",
          "https://ollie.ai/privacy-policy/"
        ],
        [
          "Compliance",
          "SOC 2 Type 1",
          "Footer states SOC 2 Type 1 compliant, with a trust center at trust.ollie.ai; legal entity Confabulation Corporation. CCPA rights with no sale/share for cross-context behavioral advertising.",
          "https://ollie.ai/"
        ]
      ],
      "setup_steps": [
        "Text Ollie's number - free tier needs no login ('Free to start. No login, just a text!').",
        "Connect Google or Outlook email and calendar for inbox/calendar awareness.",
        "Add Ollie to the family group chat - one subscription covers everyone.",
        "Manage the plan by texting: upgrade, downgrade or cancel anytime."
      ],
      "price_tags": [
        "free",
        "listed",
        "usage"
      ],
      "price_summary": "Free 50 messages/month; $25/month for 150; $100/month for 1,000 (vendor pricing)",
      "page_url": "https://agentcomparison.net/agents/ollie/"
    },
    {
      "id": "orbits",
      "name": "Orbits",
      "vendor": "Orbits",
      "kind": "product",
      "category": "household assistant with text and group chat",
      "group": "personal-assistants",
      "summary": "Orbits is a household assistant that combines shared calendars, tasks, lists, chores, meals and home upkeep in an app. Its assistant, Bit, can be texted directly or added to a family group chat, and its site says proposed changes are shown for approval before they are applied.",
      "one_liner": "A household app and text assistant (Bit) for family calendars, chores, lists and home upkeep.",
      "best_fit": "Families who want one shared place for calendars, chores, lists and home upkeep, with an assistant they can text.",
      "poor_fit": "Anyone who wants a simple texting-only assistant, or needs a published price.",
      "decision_note": "Orbits is a broader household system than a pure texting agent: the app holds the records and Bit works by text and in chat. We did not find a price on the pages we read. Phone-call bookings are described by the vendor; it says it confirms details with you before placing the call.",
      "availability": "iOS and Android apps plus text and group chat per the vendor site.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Text Bit directly or add it to a family group chat",
          "source_url": "https://tryorbits.com/",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Proposed changes shown for approval before they are applied",
          "source_url": "https://tryorbits.com/",
          "observed_at": "2026-10-03",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Text Bit directly or add it to the family group chat; creates events, reminders and coordinates shared lists, meals and chores.",
        "Connects email so Bit catches household details like school changes and appointments; turns connected email and family messages into updates.",
        "Places real phone calls for bookings after the user approves the date, time and booking name; sends the outcome back.",
        "Hands a grocery cart to Instacart for checkout."
      ],
      "limitations": [
        "No price found on the pages we read",
        "Backed by a16z Speedrun among others per TechCrunch (investor claim, not verified)",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "App (iOS, Android) plus text and family group chat"
      ],
      "privacy_security": "Orbits' privacy policy (entity Nemus AI, Inc.) states third-party model providers are prohibited from using customer content to train their models except as required to provide the inference service, and that Orbits keeps no separate 'quality,' 'training,' or 'memory' copy of personal information. Phone numbers, SMS opt-in data and message content are not shared with third parties. Connected-account data (Google, Apple iCloud, Microsoft) is deleted or de-identified per its retention section; the policy says it does not read, store or process messages from unconnected sources. Personal information is not sold, including children's. Payments run through third-party providers without full card storage. Checked 23 October 2026.",
      "pricing": {
        "amount": null,
        "currency": "USD",
        "billing_period": null,
        "allowances": null,
        "usage_charges": "Not stated",
        "verified_at": "2026-10-23",
        "source_url": "https://tryorbits.com/",
        "note": "No pricing published on tryorbits.com (checked 23 October 2026); /pricing returns a 404. Cost unconfirmed.",
        "class": "see_vendor"
      },
      "sources": [
        "https://tryorbits.com/",
        "https://tryorbits.com/privacy",
        "https://techcrunch.com/2026/10/03/all-the-ai-agents-that-can-live-in-your-text-messages/"
      ],
      "changelog": [
        {
          "date": "2026-10-03",
          "change": "Added from the TechCrunch roundup of 3 October 2026 (used as a lead only) after checking the vendor's own pages. Documentation research only."
        },
        {
          "date": "2026-10-21",
          "change": "Re-checked vendor site: no pricing published, claims unchanged."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Not published",
          "No pricing page or plan list was found on tryorbits.com as of 23 October 2026; /pricing 404s. Treat cost as unconfirmed.",
          "https://tryorbits.com/"
        ],
        [
          "Approval",
          "User approves bookings",
          "For bookings, the user picks the date, time and booking name and 'approves the details'; Bit then places the call and sends the outcome back.",
          "https://tryorbits.com/"
        ],
        [
          "Training use",
          "Providers barred",
          "Policy: third-party model providers are prohibited from using customer content to train their models except as required to provide the inference service. No separate quality/training/memory copy is retained.",
          "https://tryorbits.com/privacy"
        ],
        [
          "Messaging data",
          "Not shared",
          "Phone numbers, SMS opt-in data and message content are not shared with third parties; SMS opt-in data goes only to subcontractors delivering the service.",
          "https://tryorbits.com/privacy"
        ],
        [
          "Sale of data",
          "Not sold",
          "The policy states personal information is not sold, including information relating to children. Payments are processed by third-party providers; Nemus AI, Inc. does not store full payment card data.",
          "https://tryorbits.com/privacy"
        ]
      ],
      "setup_steps": [
        "Text Bit directly, or add Bit to the family group chat.",
        "Connect the family calendar and email so Bit catches school changes and appointments.",
        "For restaurant or appointment bookings, expect an approval step: you confirm the date, time and booking name before Bit calls.",
        "Pricing is not published - confirm cost with the vendor before relying on it."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Price not published on the pages we read",
      "page_url": "https://agentcomparison.net/agents/orbits/"
    },
    {
      "id": "pally",
      "name": "Pally",
      "vendor": "Pally",
      "kind": "product",
      "category": "Text-message assistant for connected apps and tasks",
      "group": "personal-assistants",
      "summary": "Pally is a cloud assistant you text through iMessage, RCS or Telegram. The vendor lists connected email, calendars, WhatsApp, files and work apps, plus research, reminders, calls and purchases. WhatsApp is a connected inbox, not the channel for chatting with Pally.",
      "one_liner": "A texting assistant for iMessage, RCS and Telegram, with a free core service and paid extras.",
      "best_fit": "People who want a texting assistant across connected email, calendars and files, and can review its permissions and data use.",
      "poor_fit": "Anyone requiring end-to-end encrypted processing, no model training by default, or a fresh confirmation before every action.",
      "decision_note": "The core service is free with a usage allowance that refills through the week, not a fixed 3-app free tier. Cloud Mac costs extra; invitation-only VIP is $1,000/month or $10,000/year for higher limits using the same AI models. Review saved permissions and model-training settings before connecting private accounts.",
      "availability": "Vendor lists iMessage, RCS and Telegram. SMS is a fallback; SMS-only users are directed to Telegram. WhatsApp connects through Linked Devices as an inbox, not as a Pally chat channel. Terms require age 18+.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Texting through iMessage, RCS and Telegram; connected WhatsApp inbox",
          "source_url": "https://pally.com/pricing",
          "observed_at": "2026-10-16",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Free core allowance",
          "region_restrictions": "Not fully specified on the pages reviewed"
        },
        {
          "name": "Connected email, calendars, files and work apps",
          "source_url": "https://pally.com/ai",
          "observed_at": "2026-10-16",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Only accounts you connect; iMessage inbox needs Mac app",
          "region_restrictions": "Not fully specified on the pages reviewed"
        },
        {
          "name": "Research, reminders, custom automations and calls",
          "source_url": "https://pally.com/pricing",
          "observed_at": "2026-10-16",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Shared usage allowance; not hands-on tested",
          "region_restrictions": "Not fully specified on the pages reviewed"
        },
        {
          "name": "Bookings and purchases with documented permission exceptions",
          "source_url": "https://pally.com/terms",
          "observed_at": "2026-10-16",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Exact requests and saved permissions may bypass fresh confirmation; wallet approval separate",
          "region_restrictions": "Not fully specified on the pages reviewed"
        }
      ],
      "permissions": [
        "Connects iMessage, WhatsApp, Gmail, Outlook, Slack, Google Calendar, Granola, Notion, Linear, Drive and more; morning brief and important-message alerts.",
        "Unlimited custom automations, including instant triggers ('text me when my landlord emails'); nightly memory tidy-up.",
        "Web research; completes bookings and purchases on the web, checking first on anything not already asked for or approved.",
        "Makes real phone calls: books tables, chases contractors, gets quotes. Gets its own phone number and email address when first needed.",
        "Trusted Pals: your Pally can coordinate with other people's Pallys one person at a time."
      ],
      "limitations": [
        "The vendor calls usage almost unlimited but still applies a weekly-refilling limit; a numeric allowance is not published on the pricing page. Chat remains available at the limit.",
        "Cloud Mac is a separate paid add-on; its price was not stated on the pricing page reviewed. VIP is invitation-only, not the ordinary free plan.",
        "The home page says it never messages anyone without asking; the terms allow exact user requests, saved permissions, automations, auto-send and trusted-network exchanges without separate confirmation.",
        "Documentation researched, not hands-on tested; vendor demonstrations are not verified outcomes."
      ],
      "runtime": [
        "Messaging: iMessage, RCS, Telegram",
        "Cloud service; iMessage inbox access through the Pally Mac app"
      ],
      "privacy_security": "Pally's privacy policy discloses sharing with hosting, AI, app-connection, messaging, browsing, payment and booking providers as needed to run the service. Two material points: (1) content may be used to improve and evaluate Pally's AI models unless you opt out - there is a 'Help train Pally's models' switch under Privacy and data at my.pally.com; (2) identifiers and internet activity are shared with advertising partners for cross-context behavioral advertising. Data is also shared with people or services you authorise (e.g. event hosts). Retention and account-deletion specifics were not located in the extracted text as of 23 October 2026.",
      "pricing": {
        "amount": 0,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Core service includes all connections, automations and calls under one weekly-refilling usage allowance. No card or invitation needed for the free plan.",
        "usage_charges": "Cloud Mac paid add-on, price not stated on the page reviewed. Invitation-only VIP: $1,000/month or $10,000/year, higher usage limit and the same AI models. Merchant purchases remain separate costs.",
        "verified_at": "2026-10-23",
        "source_url": "https://pally.com/pricing",
        "note": "pally.com/pricing (page updated 18 October 2026, checked 23 October): free with 'almost unlimited' usage (weekly refilling limit, warning before cap, chat keeps working). Cloud Mac is a separate paid add-on. Pally VIP is invite-only at $1,000/month or $10,000/year with higher limits and the same models. Older $25 Pro / $100 Max listings are discontinued.",
        "class": "free"
      },
      "sources": [
        "https://pally.com/",
        "https://pally.com/pricing",
        "https://pally.com/ai",
        "https://pally.com/terms",
        "https://pally.com/privacy"
      ],
      "changelog": [
        {
          "date": "2026-10-16",
          "change": "Rechecked current pricing, channel requirements, terms and privacy policy. Removed obsolete 3-app/$25 plan and Drive-only privacy wording; separated WhatsApp inbox from chat channels, documented permission exceptions and training opt-out. No hands-on tests."
        },
        {
          "date": "2026-10-03",
          "change": "Added from the TechCrunch roundup of 3 October 2026 (used as a lead only) after checking the vendor's own pages. Documentation research only."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Free",
          "Free with 'almost unlimited' usage (limit refills through the week; Pally warns before you reach it; chat keeps working). Cloud Mac is a separate paid add-on; Pally VIP is invite-only at $1,000/mo or $10,000/yr with the same models. Older $25/$100 listings are discontinued.",
          "https://pally.com/pricing"
        ],
        [
          "Channels",
          "iMessage/RCS/WhatsApp/Telegram",
          "Text Pally over iMessage or RCS, or message it on WhatsApp or Telegram; you can connect your own WhatsApp account so Pally can help in your existing chats.",
          "https://pally.com/pricing"
        ],
        [
          "Approval",
          "Checks first",
          "Completes bookings and purchases on the web, 'and checks with you first on anything you haven't already asked for or approved'.",
          "https://pally.com/pricing"
        ],
        [
          "Training use",
          "Opt-out",
          "Content may be used to improve and evaluate Pally's AI models unless you opt out; the 'Help train Pally's models' switch lives under Privacy and data at my.pally.com.",
          "https://pally.com/privacy"
        ],
        [
          "Advertising",
          "Shares for ads",
          "Identifiers and internet activity are shared with advertising partners for cross-context behavioral advertising, per the privacy policy's sharing table.",
          "https://pally.com/privacy"
        ]
      ],
      "setup_steps": [
        "Text Pally on iMessage/RCS, or message it on WhatsApp/Telegram - no card, waitlist or invitation needed.",
        "Connect your apps (Gmail, Outlook, Slack, calendar, Notion, Linear, Drive) from my.pally.com.",
        "Review the 'Help train Pally's models' switch under Privacy and data - model training is on unless you opt out.",
        "Cloud Mac is a separate paid add-on; ignore old $25/$100 listings - those plans are discontinued."
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Free core service with limits; Cloud Mac extra; invite-only VIP $1,000/month or $10,000/year",
      "page_url": "https://agentcomparison.net/agents/pally/"
    },
    {
      "id": "rene",
      "name": "Rene",
      "vendor": "Rene",
      "kind": "product",
      "category": "personal assistant in Messages, WhatsApp and Telegram",
      "group": "personal-assistants",
      "summary": "Rene is an assistant you message in Messages on iPhone, or WhatsApp or Telegram (on Android, WhatsApp or Telegram). Its site describes coordinating email, calendar, notes, reminders and work apps, doing browser tasks and making plans between your Rene and a friend's Rene.",
      "one_liner": "A text-first personal assistant that coordinates email, calendar and plans, including with friends' assistants.",
      "best_fit": "People who want one assistant to handle the small coordination work between email, calendar and friends.",
      "poor_fit": "Anyone who needs a published price or a detailed list of approval controls before connecting accounts.",
      "decision_note": "Rene's site and FAQ say data is not sold or used to train models and that email and calendar use delegated access, so it never stores your password. We did not find a price or plan list on the pages we read, so price is unknown.",
      "availability": "iPhone Messages, WhatsApp, Telegram; Android through WhatsApp or Telegram (vendor FAQ).",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Messages, WhatsApp and Telegram access",
          "source_url": "https://rene.co/faq",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Email and calendar through delegated access, not stored passwords",
          "source_url": "https://rene.co/faq",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Messaging channels: iPhone Messages app, WhatsApp and Telegram (Android: WhatsApp or Telegram).",
        "Email and calendar connect via delegated access - the vendor states Rene never sees or stores your password.",
        "Connects to existing tools (email, calendar, notes, reminders, work apps) which remain the source of truth; Rene does not replace them.",
        "FAQ: Rene 'can surface what changed, draft the next step, and ask before anything sensitive is sent or changed' - vendor statement, not tested.",
        "Rene-to-Rene coordination: two users' Renes can coordinate plans between them (vendor feature claim)."
      ],
      "limitations": [
        "No price found on the pages we read",
        "Rene-to-Rene planning needs the other person to use Rene too",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Messaging: iMessage/Messages, WhatsApp, Telegram"
      ],
      "privacy_security": "Rene's site and FAQ state: your data is never sold and never shared without permission; what you share with Rene is not used to train AI models; email and calendar use delegated access so Rene never sees or stores your password. Existing apps (email, calendar, notes, work apps) remain the source of truth. Checked 23 October 2026.",
      "pricing": {
        "amount": null,
        "currency": "USD",
        "billing_period": null,
        "allowances": null,
        "usage_charges": "Not stated",
        "verified_at": "2026-10-22",
        "source_url": "https://rene.co/",
        "note": "Rene's homepage says 'Free to start, no credit card required' (re-verified 22 October 2026). No paid-tier prices are published on the pages we read; what 'free to start' excludes is not specified. Unknown is not free.",
        "class": "see_vendor"
      },
      "sources": [
        "https://rene.co/",
        "https://rene.co/faq",
        "https://techcrunch.com/2026/10/03/all-the-ai-agents-that-can-live-in-your-text-messages/"
      ],
      "changelog": [
        {
          "date": "2026-10-23",
          "change": "Field-level audit from live site + FAQ: free to start, no card; channels Messages/WhatsApp/Telegram; not used to train AI models, never sold/shared; delegated access (no passwords); asks before sensitive sends/changes; Rene-to-Rene coordination feature. Coverage rows, permissions, setup steps, privacy fields filled. No hands-on testing."
        },
        {
          "date": "2026-10-22",
          "change": "Homepage now states 'Free to start, no credit card required'; pricing note updated (previously no price found). FAQ re-verified: iPhone Messages, WhatsApp, Telegram; delegated email/calendar access; vendor says shared data is not used to train AI models and asks before sensitive sends. No hands-on testing."
        },
        {
          "date": "2026-10-04",
          "change": "Added from https://techcrunch.com/2026/10/03/all-the-ai-agents-that-can-live-in-your-text-messages/ (used as a lead only) after checking the vendor's own pages. Documentation research only."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Free to start",
          "Homepage: 'Free to start, no credit card required.' No paid-tier pricing published as of 23 October 2026.",
          "https://rene.co/"
        ],
        [
          "Channels",
          "Messages, WhatsApp, Telegram",
          "FAQ: message Rene in the iPhone Messages app, WhatsApp, or Telegram; on Android, WhatsApp or Telegram.",
          "https://rene.co/faq"
        ],
        [
          "Training use",
          "Not used to train models",
          "Site and FAQ: what you share with Rene is not used to train AI models; data never sold, never shared without permission.",
          "https://rene.co/"
        ],
        [
          "Account security",
          "Delegated access",
          "Email and calendar use delegated access; Rene never sees or stores your password (vendor statement).",
          "https://rene.co/faq"
        ],
        [
          "Sensitive actions",
          "Asks first",
          "FAQ: Rene asks before anything sensitive is sent or changed; proactive without taking control (vendor statement).",
          "https://rene.co/faq"
        ]
      ],
      "setup_steps": [
        "Say hi to Rene on your preferred channel (Messages, WhatsApp or Telegram) and connect one app first - email or calendar - via the delegated-access grant.",
        "Let it surface and draft for a day before handing off anything sensitive; check its asks-first behaviour yourself.",
        "Keep your existing apps as the source of truth and review what context you have connected before inviting Rene-to-Rene coordination."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Price not published on the pages we read",
      "page_url": "https://agentcomparison.net/agents/rene/"
    },
    {
      "id": "town",
      "name": "Town",
      "vendor": "Town",
      "kind": "product",
      "category": "AI assistant (Townie) for professional work",
      "group": "work-business",
      "summary": "Town gives each user an assistant, a 'Townie', that works with email, calendar, documents, Slack and other tools. Its docs describe searching the web, reading email, checking calendars, drafting replies, creating documents, running code and recurring routines, and reaching it from Telegram or WhatsApp (inbound only).",
      "one_liner": "An assistant for email, calendar, documents and Slack with approval modes and recurring routines.",
      "best_fit": "Professionals who want routines such as meeting briefings and email triage with explicit approval settings.",
      "poor_fit": "Anyone who wants a purely consumer texting assistant; Town is built around work tools.",
      "decision_note": "Town's docs explain three approval modes: require approval for all writes, tool defaults, or autonomous. Calendar changes need approval by default. Credits set usage: a free plan with 30 chat messages a month, then $15 a month and $49 a month plans on the pricing page.",
      "availability": "Web, Telegram and WhatsApp (inbound only) per the vendor site. Region details not checked.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Approval modes for write actions",
          "source_url": "https://www.town.com/docs/using-town/assistant",
          "observed_at": "2026-10-04",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Recurring routines such as meeting briefings and email triage",
          "source_url": "https://www.town.com/",
          "observed_at": "2026-10-04",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Telegram and WhatsApp access, inbound only",
          "source_url": "https://www.town.com/",
          "observed_at": "2026-10-04",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Town's docs list three approval modes. 'Require approval' needs your explicit approval for every write action; 'Autonomous' runs all tool calls without approval; the default depends on each tool. Calendar changes require approval by default."
      ],
      "limitations": [
        "Free plan: 30 chat messages a month and a 14-day Pro trial",
        "Telegram and WhatsApp are inbound-only; the Townie cannot read or send messages there (vendor)",
        "Credits determine how much work you can run; heavier plans cost more",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Web app; Telegram and WhatsApp messaging inbound"
      ],
      "privacy_security": "Town's site says it does not share, sell or train on your data. We did not review the privacy policy or any audit report.",
      "pricing": {
        "amount": 15,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Free: 30 chat messages a month. Starter: 1,250 credits. Pro: 5,000 credits at $49 a month",
        "usage_charges": "Pro has pay-as-you-go overage; higher Power tiers also listed",
        "verified_at": "2026-10-22",
        "source_url": "https://www.town.com/pricing",
        "note": "town.com/pricing lists a free plan, Starter at $15 a month, Pro at $49 a month, and Power and Power Plus tiers with larger credit pools. Prices for the top tiers were not captured here."
      },
      "sources": [
        "https://www.town.com/",
        "https://www.town.com/docs/using-town/assistant",
        "https://www.town.com/pricing",
        "https://techcrunch.com/2026/10/03/all-the-ai-agents-that-can-live-in-your-text-messages/"
      ],
      "changelog": [
        {
          "date": "2026-10-04",
          "change": "Added from https://techcrunch.com/2026/10/03/all-the-ai-agents-that-can-live-in-your-text-messages/ (used as a lead only) after checking the vendor's own pages. Documentation research only."
        }
      ],
      "price_tags": [
        "free",
        "listed",
        "usage"
      ],
      "price_summary": "Free plan; Starter $15/month; Pro $49/month; credit based (vendor pricing)",
      "page_url": "https://agentcomparison.net/agents/town/"
    },
    {
      "id": "lindy",
      "name": "Lindy",
      "vendor": "Lindy",
      "kind": "product",
      "category": "AI teammate in Slack",
      "group": "work-business",
      "summary": "Lindy describes an AI teammate that lives in Slack and knows what your company knows. Everyone gets a private Lindy in DMs and a shared one in channels. Its docs list answering questions across tools, multi-step actions and creating decks, sheets and reports.",
      "one_liner": "An AI teammate in Slack with integration write guardrails whose scope depends on channel and settings.",
      "best_fit": "Teams using shared Slack threads who can configure integration write guardrails and review which account each connection represents.",
      "poor_fit": "Solo users who do not use Slack, or teams that need a free plan beyond the 7-day trial credits.",
      "decision_note": "Current pricing starts at $49.99 per active user per month for 5,000 credits, with top-ups at $10 per 1,000. The pricing FAQ says outside-impact actions always wait for approval, but the integration documentation limits write guardrails to shared Slack threads and allows Always allow. Do not assume every channel or account has a human approval gate.",
      "availability": "Slack-based; the pricing page also lists iMessage with your assistant. Region details not checked.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Slack teammate with private DMs and shared channel use",
          "source_url": "https://docs.lindy.ai/",
          "observed_at": "2026-10-14",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Configurable write guardrails in shared Slack threads",
          "source_url": "https://docs.lindy.ai/integrations/overview",
          "observed_at": "2026-10-14",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Integration docs list Always allow (writes without asking), Require approval (asks in Slack) and Don't offer (no writes with the tool). Reads are not guarded. These write guardrails apply only in shared Slack threads, not web chat, Slack DMs, iMessage or SMS.",
        "The pricing FAQ still says outside-impact actions wait for approval. That conflicts with the narrower channel/settings scope in the integration docs; verify the actual configuration before relying on approval."
      ],
      "limitations": [
        "Free credits last 7 days; paid usage starts at $49.99 per user per month for 5,000 credits.",
        "Every active workspace user is billed a seat",
        "Monthly credits do not roll over",
        "Described by the vendor; not tested by AgentComparison",
        "Write guardrails cover shared Slack threads, not web chat, Slack DMs, iMessage or SMS, per integration docs.",
        "The pricing FAQ makes a broader approval claim than the integration docs. A guardrail applies across all accounts of an integration, not separately per account."
      ],
      "runtime": [
        "Slack; iMessage listed on the pricing page"
      ],
      "privacy_security": "The privacy notice, dated 28 September 2026, lists cloud, AI, integration and communication providers. It excludes Google API data from AI model training or evaluation and says personal information will not train a generative AI product. It retains data as needed for the stated purposes and legal requirements, with deletion or anonymization when no longer needed; backups may be isolated until deletion is possible. It does not give a single fixed deletion deadline. These are policy statements, not an independent security audit.",
      "pricing": {
        "amount": 49.99,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "5,000 credits per user per month; $50 in free credits for 7 days on signup, no card",
        "usage_charges": "Top-ups $10 per 1,000 credits; credits pause when exhausted",
        "verified_at": "2026-10-23",
        "source_url": "https://www.lindy.ai/pricing",
        "note": "Pricing page re-verified 23 October 2026: Team is now from $29.99 per active user/month for 3,000 credits each (previously $49.99/5,000 - cut between 22 and 23 October). Free tier: $50 in credits lasting 7 days, no card. Top-ups remain $10 per 1,000 credits, pooled workspace; credits do not roll over; Lindy pauses credit-using actions at the cap rather than surprise-billing. Slack joiners get a 7-day free trial before the seat is billed; direct signups are billed immediately."
      },
      "sources": [
        "https://docs.lindy.ai/",
        "https://www.lindy.ai/pricing",
        "https://docs.lindy.ai/integrations/overview",
        "https://docs.lindy.ai/teammate/setup",
        "https://www.lindy.ai/privacy"
      ],
      "changelog": [
        {
          "date": "2026-10-16",
          "change": "Expanded connection identity, channel-specific guardrails and approver scope with eight evidence rows, setup guidance and privacy/retention notes. Preserved the pricing-FAQ conflict. No hands-on tests."
        },
        {
          "date": "2026-10-14",
          "change": "Corrected current Team price and credit allowance; replaced blanket approval wording with documented channel/settings scope and the unresolved pricing-FAQ conflict. No hands-on testing."
        },
        {
          "date": "2026-10-04",
          "change": "Added from the user-supplied shortlist (used as a lead only) after checking the vendor's own pages. Documentation research only."
        },
        {
          "date": "2026-10-23",
          "note": "Fresh source-to-field re-check (phase 2): Team pricing cut from $49.99/5,000 credits to $29.99/3,000 credits per active user; top-up, pause-at-cap, free-credit and Slack-trial terms re-verified. Record counts toward the detailed ledger again."
        }
      ],
      "coverage_rows": [
        [
          "Shared and private work",
          "Documented",
          "Channel mentions continue inside the thread. DMs have the fuller private toolset; shared-channel answers have narrower public-safe scope.",
          "https://docs.lindy.ai/teammate/setup"
        ],
        [
          "Personal and team connections",
          "Identity matters",
          "Owners/admins can share a connection. Everyone then acts with that connected account's permissions, not necessarily their own.",
          "https://docs.lindy.ai/integrations/overview"
        ],
        [
          "Write guardrails",
          "Shared Slack threads only",
          "Always allow, Require approval and Don't offer govern writes. Web chat, Slack DMs, iMessage and SMS are outside these guardrails.",
          "https://docs.lindy.ai/integrations/overview"
        ],
        [
          "Read access",
          "Not guarded",
          "Reads are not approval-gated by these write guardrails. Connecting a tool enables all its actions; inspect the actual account scope.",
          "https://docs.lindy.ai/integrations/overview"
        ],
        [
          "Who can approve",
          "Documented",
          "Requester or workspace owner/admin can approve an action. Always allow changes future behavior, with scope depending on who chooses it.",
          "https://docs.lindy.ai/integrations/overview"
        ],
        [
          "Pricing and seats",
          "Published, not a usage forecast",
          "From $29.99 per active user/month for 3,000 credits each, pooled (cut from $49.99/5,000 before 23 Oct 2026). Top-ups $10/1,000; monthly credits do not roll over; Lindy pauses credit-using actions at the cap.",
          "https://www.lindy.ai/pricing"
        ],
        [
          "Approval claim conflict",
          "Unresolved in vendor pages",
          "Pricing FAQ says all outside-impact actions await approval, while integration docs limit guardrails by channel and allow silent writes. Do not rely on the broad FAQ alone.",
          "https://docs.lindy.ai/integrations/overview"
        ],
        [
          "Privacy and retention",
          "Policy, not tested",
          "Policy excludes Google API training/evaluation and gives purpose/legal retention criteria, not a universal deletion deadline.",
          "https://www.lindy.ai/privacy"
        ]
      ],
      "setup_steps": [
        "Review active-seat billing and the seven-day trial limit before inviting or mentioning teammates.",
        "Connect only the account needed for a small task. Keep it personal unless you intend the team to use that account's permissions.",
        "If using shared Slack threads, choose Require approval or Don't offer for relevant integrations, and check who can change the rule.",
        "Do not assume those guardrails apply in DMs, web chat, iMessage or SMS. Start with a read-only task and check the actual source and account.",
        "Review recorded actions, credit use and connection health before increasing scope; disconnect unused connections."
      ],
      "price_tags": [
        "listed",
        "usage"
      ],
      "price_summary": "From $49.99/user/month for 5,000 credits; 7-day free credits (vendor pricing)",
      "page_url": "https://agentcomparison.net/agents/lindy/"
    },
    {
      "id": "kiro",
      "name": "Kiro",
      "vendor": "Kiro",
      "kind": "product",
      "category": "spec-driven coding agent and IDE",
      "group": "coding-agents",
      "summary": "Kiro turns prompts into requirements, design and sequenced tasks, then implements them with parallel agents locally or in the cloud. Its site also describes property-based tests, a headless CLI for CI and model choice including Claude, GPT and open-weight models.",
      "one_liner": "A coding agent that turns prompts into specs and tasks, then builds with parallel agents.",
      "best_fit": "Developers who want specs and checks before the agent writes code.",
      "poor_fit": "Anyone who needs fixed monthly pricing; usage is credit-based and heavier models cost more credits.",
      "decision_note": "Kiro combines specs and code work with credit-based pricing. The useful control question is which surface and version you use: current IDE/CLI capability rules differ from Web sandboxing and older trust settings. Individual paid accounts do not automatically get the enterprise service-improvement exclusion.",
      "availability": "Hosted and local; enterprise plan on request (vendor pricing page).",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Specs, tasks and parallel agents",
          "source_url": "https://kiro.dev/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Credit tiers from free to $200 a month",
          "source_url": "https://kiro.dev/pricing",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Current IDE and CLI documentation uses capability rules with deny, ask and allow effects. Default policy allows workspace reads and common read-only commands; other operations prompt, with protected settings and workspace-trust rules. Web uses an isolated cloud sandbox instead of the same per-action prompt model.",
        "IDE 1.x replaces the older trustedCommands model and CLI V3 replaces CLI 2.x tool settings. Older security guidance still describes prefix trust and supervised file review. Check your installed version; supervised review is not a sandbox or a universal access boundary."
      ],
      "limitations": [
        "Model and credit availability vary by plan and country; add-on credits cost $0.04 each on the pricing page. No promotional credit eligibility is assumed.",
        "Permissions depend on surface and version. Web does not use the local IDE/CLI interactive permission model.",
        "Current and older security pages use different approval descriptions; check installed-version documentation and actual rules. Supervised mode is not an isolation boundary.",
        "Data-sharing opt-out does not remove abuse-detection retention. Storage location and inference location are different questions.",
        "Vendor documentation only; no code-quality, command-safety or task-success tests run."
      ],
      "runtime": [
        "IDE, CLI and cloud sessions per the vendor site"
      ],
      "privacy_security": "Re-verified 22 October 2026 against Kiro's data-protection docs. Free Tier and individual subscribers (paid via social login or AWS Builder ID): your content - questions, inputs, and the responses and code Kiro generates - may be used for service improvement including model training, subject to your opt-out (IDE: uncheck Data Sharing and Prompt Logging > Content Collection for Service Improvement; telemetry has a separate checkbox). Free-tier inputs may also be stored up to 60 days for abuse detection. Enterprise users: content is NOT used for service improvement; admin-enabled prompt logging/usage reporting goes to a customer-owned S3 bucket. Model-specific retention requirements for abuse detection may apply across tiers. Vendor documentation, not independently audited.",
      "pricing": {
        "amount": 20,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Free: 50 credits. Paid tiers: monthly credit allowance shown in the Kiro subscription dashboard (not published on the public FAQ).",
        "usage_charges": "Credits metered per prompt (model multipliers apply; Sonnet 4.6 = 1.3x Auto per vendor example); add-on credits $0.04 each.",
        "verified_at": "2026-10-26",
        "source_url": "https://kiro.dev/pricing",
        "note": "kiro.dev/pricing read 26 October 2026: Free tier $0 (50 credits), Pro $20, Pro+ $40, Pro Max $100, Power $200 per user/month; add-on credits $0.04 (packs $5-$100, 12-month rollover); monthly plan credits do not roll over; Enterprise via AWS with optional $0.04/credit overage. Prices exclude tax; GovCloud ~20% higher, no free tier.",
        "class": "amount"
      },
      "sources": [
        "https://kiro.dev/",
        "https://kiro.dev/pricing",
        "https://kiro.dev/docs/permissions/",
        "https://kiro.dev/docs/ide/whats-new-v1/permissions/",
        "https://kiro.dev/docs/cli/v3/permissions/",
        "https://kiro.dev/docs/privacy-and-security/data-protection/",
        "https://kiro.dev/docs/privacy-and-security/"
      ],
      "changelog": [
        {
          "date": "2026-10-16",
          "change": "Added version- and surface-scoped permissions, service-improvement and abuse-retention distinctions, eight evidence rows and a narrow setup path. No hands-on tests."
        },
        {
          "date": "2026-10-04",
          "change": "Added from the user-supplied shortlist (used as a lead only) after checking the vendor's own pages. Documentation research only."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: data-protection docs re-read - opt-out-based content use for service improvement (incl. model training) on Free/individual tiers, enterprise excluded, 60-day abuse-detection storage for free tier, IDE opt-out path documented. Legal pages live under kiro.dev/docs, not kiro.dev/privacy."
        },
        {
          "date": "2026-10-26",
          "change": "Pricing field audit: class set to amount ($20 Pro entry), tiers/add-on/rollover rules re-verified on kiro.dev/pricing 26 Oct. checked_label unchanged (field-level audit, not a full-record check)."
        }
      ],
      "coverage_rows": [
        [
          "Specs and implementation",
          "Documented",
          "The vendor describes requirements, design and tasks before implementation. This is a workflow claim, not proof of correctness.",
          "https://kiro.dev/"
        ],
        [
          "IDE and CLI permission rules",
          "Documented, version-scoped",
          "Current rules use deny, ask and allow per capability. Workspace reads and some read-only commands are allowed by default; other operations prompt.",
          "https://kiro.dev/docs/permissions/"
        ],
        [
          "Migration from older trust settings",
          "Documented",
          "IDE 1.x replaces 0.x trustedCommands. CLI V3 replaces 2.x tool settings; verify converted rules rather than assuming old behavior.",
          "https://kiro.dev/docs/cli/v3/permissions/"
        ],
        [
          "Web sessions",
          "Different control model",
          "Web uses a managed cloud sandbox, not permissions.yaml or the same per-action approval prompts.",
          "https://kiro.dev/docs/permissions/"
        ],
        [
          "Supervised review",
          "Not a security boundary",
          "Older security guidance describes file review and revert, not restricted reads or network access. Current autonomy wording differs; verify version and configuration.",
          "https://kiro.dev/docs/privacy-and-security/"
        ],
        [
          "Data use by account type",
          "Documented distinction",
          "Free and individual content may support service improvement/model training unless opted out; enterprise content is excluded from service improvement.",
          "https://kiro.dev/docs/privacy-and-security/data-protection/"
        ],
        [
          "Abuse retention and regions",
          "Not zero retention",
          "Free inputs may remain up to 60 days for abuse detection; model-specific retention can apply more widely. Inference region can differ from storage.",
          "https://kiro.dev/docs/privacy-and-security/data-protection/"
        ],
        [
          "Credit pricing",
          "Published, not a bill estimate",
          "Free 50 credits; paid tiers $20/$40/$100/$200 per user per month for 1,000/2,000/5,000/10,000 credits. Add-ons $0.04 each, taxes extra.",
          "https://kiro.dev/pricing"
        ]
      ],
      "setup_steps": [
        "Choose IDE, CLI or Web and record the installed version before following permission instructions.",
        "Use a disposable repository without production credentials. Review the default policy and keep any added allowances narrow.",
        "For local IDE/CLI, review capability rules and workspace trust. Do not copy older prefix-trust settings blindly.",
        "For Free or individual accounts, review content-sharing and telemetry settings separately; read the abuse-retention exceptions before supplying private code.",
        "Try a small change with clear acceptance tests, inspect the diff and commands, and check credit use before increasing scope."
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Free 50 credits; Pro $20/month for 1,000 credits; up to $200/month (vendor pricing)",
      "page_url": "https://agentcomparison.net/agents/kiro/"
    },
    {
      "id": "github-copilot-cloud-agent",
      "name": "GitHub Copilot cloud agent",
      "vendor": "GitHub (Microsoft)",
      "kind": "product",
      "category": "coding agent on GitHub",
      "group": "coding-agents",
      "summary": "GitHub's docs say Copilot cloud agent can research a repository, make an implementation plan and change code on a branch, then let you review the diff, iterate and create a pull request. It works in its own ephemeral environment powered by GitHub Actions, where it can run tests and linters.",
      "one_liner": "GitHub's background coding agent that plans, edits a branch and opens pull requests.",
      "best_fit": "Teams already on GitHub that can split work into single-repository tasks, review generated changes and run checks before a human merges the pull request.",
      "poor_fit": "Cross-repository changes in one task, non-GitHub hosting, or teams expecting the agent to merge changes without a human. Also a poor fit if setup steps or MCP access cannot be reviewed before granting access.",
      "decision_note": "A good fit for bounded tasks in a GitHub repository with clear acceptance tests and a reviewer. Distinguish cloud agent from local IDE agent mode: cloud work runs in an Actions-powered environment and uses both AI credits and Actions minutes. Before enabling it, check repository policy, workflow approval, firewall scope and MCP access. A firewall is not a guarantee that every process is isolated.",
      "availability": "Available on paid Copilot plans when enabled. Business and Enterprise users need an administrator to enable the relevant policy; repository owners can opt repositories out. Managed-user personal repositories are not supported. Planning and iteration before a PR are documented on GitHub.com, with Teams and Slack in public preview; other integrations may create a PR directly.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Research, plan and change one GitHub repository in an Actions-powered environment",
          "source_url": "https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Paid Copilot plan and relevant policy enabled",
          "region_restrictions": "Not verified"
        },
        {
          "name": "Human PR review and restricted branch writes",
          "source_url": "https://docs.github.com/en/copilot/concepts/agents/cloud-agent/risks-and-mitigations",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Paid Copilot plan and relevant policy enabled",
          "region_restrictions": "Not verified"
        },
        {
          "name": "Default firewall with explicit MCP and setup-step exclusions",
          "source_url": "https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent/customize-the-agent-firewall",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Paid Copilot plan and relevant policy enabled",
          "region_restrictions": "Not verified"
        },
        {
          "name": "Default code, dependency and secret checks before completing work",
          "source_url": "https://docs.github.com/en/copilot/concepts/agents/cloud-agent/risks-and-mitigations",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "high",
          "plan_restrictions": "Paid Copilot plan and relevant policy enabled",
          "region_restrictions": "Not verified"
        }
      ],
      "permissions": [
        "Only users with repository write access can trigger the agent. It writes to one branch: the existing PR branch for an @copilot PR task, otherwise a new copilot/ branch. Branch protections and required checks still apply.",
        "GitHub documents human review and merge: the agent cannot approve or merge a pull request or mark its own draft ready for review. The user who requested the PR cannot supply its required approval.",
        "By default, PR Actions workflows wait for a person with write access to click Approve and run workflows. Administrators can allow automatic runs, so confirm the actual repository policy before relying on that gate.",
        "The default firewall covers processes the agent starts through its Bash tool inside the Actions appliance. It does not cover MCP servers or configured setup-step processes, and GitHub warns that sophisticated attacks may bypass it. Inspect allowed domains, setup scripts and MCP access separately.",
        "Protect Copilot and MCP configuration files with CODEOWNERS and review rules. GitHub distinguishes ordinary Actions secrets from Agents secrets and variables deliberately provided to the cloud agent; grant only the data a task needs."
      ],
      "limitations": [
        "Cloud agent requires a paid Copilot plan; the Free plan is not a free cloud-agent tier",
        "Uses both GitHub AI credits and GitHub Actions minutes; subscription price alone is not a per-task cost",
        "One repository and one branch per task; one pull request; maximum session duration 59 minutes",
        "MCP settings can broaden repository context; default firewall does not cover MCP servers or setup steps",
        "Branch rules can block the agent; do not remove review requirements merely to make a task run",
        "Plans table total credits and Max FAQ base-credit wording differ; variable Flex allotments can change. Verify the actual account allowance.",
        "Vendor documentation, not a hands-on test or security audit"
      ],
      "runtime": [
        "Ephemeral development environment powered by GitHub Actions; not the same as local IDE agent mode",
        "One repository, one branch and one pull request per task; session maximum 59 minutes",
        "Repository context by default; MCP configuration can broaden access"
      ],
      "privacy_security": "Re-verified 22 October 2026 against GitHub's responsible-use docs for Copilot agents: the cloud agent runs in an ephemeral development environment with a network firewall ENABLED BY DEFAULT that blocks outbound connections to unauthorised hosts against code/data exfiltration; automated security scanning runs on its output; work lands as pull requests for human review and iteration. The doc does not state a prompt/code retention period for the cloud agent. Vendor documentation, not independently audited.",
      "pricing": {
        "amount": 10,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Plans table lists Pro $15, Pro+ $70 and Max $200 total monthly credits, including variable Flex allotments. FAQ says Max includes $100/month; discrepancy remains.",
        "usage_charges": "Cloud agent consumes AI credits (model and token dependent) plus GitHub Actions minutes. Included allowances can avoid additional charges; check both budgets.",
        "verified_at": "2026-10-22",
        "source_url": "https://github.com/features/copilot/plans",
        "note": "Individual subscriptions list Pro $10, Pro+ $39 and Max $100 per user/month. The plans table lists $15 / $70 / $200 total monthly credits, including variable Flex allotments, while its FAQ says Max includes $100/month. Verify actual account allowances rather than treating either as guaranteed. Cloud agent requires a paid plan and also uses Actions minutes. Organization/enterprise additional AI-credit usage is enabled by default unless disabled; this is not an individual-plan assumption."
      },
      "sources": [
        "https://docs.github.com/en/copilot/concepts/copilot-surfaces/copilot-on-github",
        "https://github.com/features/copilot/plans",
        "https://docs.github.com/en/copilot/concepts/security-governance-and-network-settings/risks-and-mitigations",
        "https://docs.github.com/en/copilot/how-tos/copilot-on-github/customize-copilot/customize-the-firewall",
        "https://docs.github.com/en/enterprise-cloud@latest/copilot/tutorials/cloud-agent/build-guardrails",
        "https://docs.github.com/en/copilot/concepts/billing-and-usage/organizations-and-enterprises/billing",
        "https://docs.github.com/en/copilot/how-tos/use-copilot-agents/cloud-agent/troubleshoot-cloud-agent",
        "https://docs.github.com/en/copilot/responsible-use/agents"
      ],
      "changelog": [
        {
          "date": "2026-10-13",
          "change": "Added eight source-linked feature and limit rows plus a narrow start checklist from current vendor documentation. No hands-on test or ranking."
        },
        {
          "date": "2026-10-08",
          "change": "Checked seven GitHub sources. Added branch/review gates, firewall exclusions, one-repository and 59-minute limits, plus Actions-minute costs. Removed the Free price filter tag because Free does not include this cloud agent."
        },
        {
          "date": "2026-10-04",
          "change": "Added from the user-supplied shortlist (used as a lead only) after checking the vendor's own pages. Documentation research only."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: responsible-use doc re-read - ephemeral firewalled environment (default-on), automated security scanning, PR-based human review all confirmed verbatim; no retention period stated."
        }
      ],
      "coverage_rows": [
        [
          "Runtime",
          "Cloud, not IDE mode",
          "An ephemeral GitHub Actions environment can research, plan and change code. Local IDE agent mode has a different execution environment.",
          "https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent"
        ],
        [
          "Planning entry points",
          "Channel-dependent",
          "Research/plan/iterate before a PR works on GitHub.com; Teams/Slack support is public preview. Other integrations can create a PR directly.",
          "https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent"
        ],
        [
          "Branch and merge",
          "Human review",
          "Only repository writers can trigger it. It writes one branch and cannot approve, merge or mark its draft ready. The requesting user cannot supply required approval.",
          "https://docs.github.com/en/copilot/concepts/agents/cloud-agent/risks-and-mitigations"
        ],
        [
          "Workflow approval",
          "Default can change",
          "PR workflows wait for Approve and run workflows by default; administrators can allow automatic runs. Verify the actual policy.",
          "https://docs.github.com/en/copilot/concepts/agents/cloud-agent/risks-and-mitigations"
        ],
        [
          "Firewall coverage",
          "Limited boundary",
          "Firewall covers agent Bash processes in the Actions appliance, not MCP servers or setup-step processes. GitHub warns sophisticated attacks may bypass it.",
          "https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent/customize-the-agent-firewall"
        ],
        [
          "Code checks",
          "Defaults, not guarantees",
          "CodeQL, dependency advisory checks and secret scanning are documented defaults that can be disabled. Review output yourself; checks do not prove safe code.",
          "https://docs.github.com/en/copilot/concepts/agents/cloud-agent/risks-and-mitigations"
        ],
        [
          "Task bounds",
          "One repository",
          "One repository, one branch and one PR per task; maximum execution time is 59 minutes. Split larger work without removing review requirements.",
          "https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent"
        ],
        [
          "Usage costs",
          "Two budgets",
          "Cloud agent is on paid Copilot plans and consumes AI credits plus Actions minutes. Subscription price is not a per-task total; check both allowances.",
          "https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent"
        ]
      ],
      "setup_steps": [
        "Use one repository with a reviewer and clear acceptance tests; check paid-plan and repository policy.",
        "Review setup steps, MCP scope, firewall allowlist and configuration-file ownership.",
        "Set AI-credit and Actions-minute budgets and start one task under the session limit.",
        "Check diff/security results and actual workflow approval policy before a human reviews and merges."
      ],
      "price_tags": [
        "listed",
        "usage"
      ],
      "price_summary": "Cloud agent: paid plans from $10/user/month; AI credits plus Actions minutes. Free plan does not include cloud agent.",
      "page_url": "https://agentcomparison.net/agents/github-copilot-cloud-agent/"
    },
    {
      "id": "lovable",
      "name": "Lovable",
      "vendor": "Lovable",
      "kind": "product",
      "category": "AI app builder",
      "group": "coding-agents",
      "summary": "Lovable describes a full-stack AI development platform for building, iterating on and deploying web applications from natural language, including frontend, backend, database and authentication. Work happens in shared workspaces that pool credits.",
      "one_liner": "Builds full web apps from plain-language prompts, with credits and a free daily grant.",
      "best_fit": "People who want a working web app from a description and are fine reviewing and owning the generated code.",
      "poor_fit": "Anyone who needs a fixed monthly price or deep control over infrastructure; credits are consumed by messages and hosting.",
      "decision_note": "Lovable's docs say you own the code and projects you build, and that the free plan includes 5 build credits a day (up to 30 a month). Paid plan prices were not on the pages we captured, so no monthly price is listed here.",
      "availability": "Hosted web platform. Region details not checked.",
      "last_checked": "2026-10-04",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Builds web apps from natural language, with frontend, backend and authentication",
          "source_url": "https://docs.lovable.dev/introduction/welcome",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Free plan grants 5 build credits a day",
          "source_url": "https://lovable.dev/pricing",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Model-training opt-out in account settings, any plan, free (re-verified 22 October 2026).",
        "Integrations (GitHub, Supabase, MCP connectors) are per-connection with the scopes you grant; messaging/collaboration integration data is excluded from training.",
        "Privacy settings cover ad-sharing opt-out (Do Not Sell link + GPC honoured) and session-replay consent."
      ],
      "limitations": [
        "Credits are consumed by building, hosting and AI features; unused monthly credits expire after two months (vendor FAQ)",
        "Paid plan prices not captured in our check",
        "Business credits cost more than Pro credits (vendor FAQ)",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Hosted web platform with shared workspaces"
      ],
      "privacy_security": "Privacy policy (effective and last updated 15 September 2026, read 22 October 2026) is unusually explicit. Lovable trains, develops and fine-tunes its AI models on Customer Content and Usage Data BY DEFAULT, with a free opt-out in account settings on any plan; opt-out works going forward and does not retract already-assembled training data or trained models. It does NOT train on: your account and billing details; Your Users' Data (your app's own end-user data, held in your project's database); or data from messaging/collaboration integrations, which is used only to operate the integration. Trained team members may review content for model quality under confidentiality safeguards. Third-party model providers are contractually restricted from training on your content. Session-replay sampling masks project-chat input and is not used for training; consent-based where required, and Global Privacy Control is honoured. Vendor policy, not independently audited.",
      "pricing": {
        "amount": null,
        "currency": "USD",
        "billing_period": null,
        "allowances": "Free: 5 build credits a day (up to 30 a month), 20 monthly Cloud credits",
        "usage_charges": "Paid plans use a credit balance; Plan Mode costs 1 credit per message",
        "verified_at": "2026-10-04",
        "source_url": "https://lovable.dev/pricing",
        "note": "lovable.dev/pricing describes a free plan and credit-based paid plans (Pro, Business, Enterprise). The dollar amounts were not captured, so none is stated."
      },
      "sources": [
        "https://docs.lovable.dev/introduction/welcome",
        "https://lovable.dev/pricing",
        "https://lovable.dev/privacy",
        "https://lovable.dev/terms"
      ],
      "changelog": [
        {
          "date": "2026-10-04",
          "change": "Added from the user-supplied shortlist (used as a lead only) after checking the vendor's own pages. Documentation research only."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: privacy policy (15 September 2026) read - default-on model training with free any-plan opt-out (forward-looking), explicit no-training list (billing, Your Users' Data, messaging integrations), team quality review, provider training restrictions, masked session replay. Permissions rewritten."
        }
      ],
      "price_tags": [
        "free",
        "usage"
      ],
      "price_summary": "Free plan (5 build credits a day); paid plans credit based, prices not captured",
      "page_url": "https://agentcomparison.net/agents/lovable/"
    },
    {
      "id": "google-antigravity",
      "name": "Google Antigravity",
      "vendor": "Google",
      "kind": "product",
      "category": "agentic development platform",
      "group": "coding-agents",
      "summary": "Google Antigravity is Google's agentic development platform. The vendor site describes four surfaces: a desktop app for running several local agents in parallel and grouping chats into Projects, a terminal CLI, an IDE with an agent manager and artifacts, and a Python SDK for building custom agents on the same harness.",
      "one_liner": "Google's agent-first dev platform: desktop app, CLI, IDE and SDK, with a free individual plan.",
      "best_fit": "Developers who want to try an agent-first IDE and CLI without paying, within weekly rate limits.",
      "poor_fit": "Anyone who needs a known price for heavier use: the Google AI Pro price is not stated on the pricing page.",
      "decision_note": "Antigravity has a $0 individual plan with basic weekly rate limits and access to several Gemini, Claude and open-weight models. The pricing page lists Google AI Pro with higher limits and a credit pool, but states no price for it, and a Google Cloud organisation plan with consumption-based API pricing.",
      "availability": "Desktop on macOS, Windows and Linux; organisation access through Google Cloud (vendor pages).",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Parallel local agents, CLI, IDE and SDK",
          "source_url": "https://antigravity.google/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Free individual plan with weekly rate limits",
          "source_url": "https://antigravity.google/pricing",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "You are solely responsible for AI Agent actions, for authorising an agent's access to data/apps/systems, and for supervision in production (Additional ToS section 4, read 22 October 2026).",
        "Interactions collection can be changed in settings; deletion via antigravity-support@google.com.",
        "Selecting a third-party or open-source model subjects you to that model's terms (Anthropic commercial terms named explicitly).",
        "Using third-party software to access the Service (the terms name OpenClaw with Antigravity OAuth as an example) is a breach that can suspend Antigravity and Gemini CLI accounts."
      ],
      "limitations": [
        "Free plan has basic weekly rate limits; exact limits not stated on the pricing page",
        "Google AI Pro price not stated on the pricing page",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Desktop app, CLI, IDE and Python SDK per the vendor site"
      ],
      "privacy_security": "Google Antigravity Additional Terms of Service (read 22 October 2026) plus the universal Google Terms and Privacy Policy govern. Google records your usage 'Interactions' (interaction data, metadata, feedback) while the Service runs and uses them to 'evaluate, develop, and improve Google and Alphabet research, products, services and machine learning technologies'; Google employees and contractors may review them. You can change this in settings, and deletion of Interactions can be requested via antigravity-support@google.com - use continues under the Agreement until you request deletion. Enterprise access via Gemini Enterprise/Google Cloud is governed by the organisation's own terms instead. Vendor terms, not independently audited.",
      "pricing": {
        "amount": 0,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Individual plan $0 with basic weekly rate limits; Google AI Pro has more generous limits and a flexible credit pool (price not stated)",
        "usage_charges": "Organisation plan: consumption-based API pricing via Google Cloud",
        "verified_at": "2026-10-22",
        "source_url": "https://antigravity.google/pricing",
        "note": "antigravity.google/pricing lists an Individuals plan at $0/month, Google AI Pro without a price, and an organisation plan via Google Cloud."
      },
      "sources": [
        "https://antigravity.google/",
        "https://antigravity.google/pricing",
        "https://antigravity.google/terms"
      ],
      "changelog": [
        {
          "date": "2026-10-04",
          "change": "Added from the backlog after checking the vendor's own pages. Documentation research only."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: Additional ToS read - Interactions used to improve Google/Alphabet ML technologies with settings control and email deletion; user-solely-responsible agent clause; third-party harness use (OpenClaw+OAuth example) explicitly a breach; Anthropic commercial terms apply when selected."
        }
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Free individual plan with weekly limits; Google AI Pro price not stated (vendor pricing)",
      "page_url": "https://agentcomparison.net/agents/google-antigravity/"
    },
    {
      "id": "hedwig",
      "name": "Hedwig",
      "vendor": "Hedwig",
      "kind": "product",
      "category": "AI email and calendar workspace",
      "group": "work-business",
      "summary": "Hedwig puts Gmail and Google Calendar on an infinite canvas of frames, piles and sticky notes, with an AI agent that works across the board. The vendor site lists catch-up summaries, drafted replies, follow-up chasing, sorting threads into piles, finding free time and adding todos.",
      "one_liner": "Email and calendar on a canvas, with an agent that drafts replies and waits for your approval.",
      "best_fit": "Gmail and Google Calendar users who want to rearrange their inbox into boards and have an agent draft replies for approval.",
      "poor_fit": "Anyone not on Gmail and Google Calendar, or who needs a known long-term price: the Teams tier has no price yet.",
      "decision_note": "The vendor says Hedwig is free while it is new: its pricing section lists a Free plan at $0 with every feature, unlimited accounts and boards and the agent included, and a Teams tier as coming soon with no price. Free access is described as an early-user offer, not a permanent commitment beyond what the page says.",
      "availability": "Open to everyone at launch per the vendor home page; connects to Gmail and Google Calendar.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Gmail and Google Calendar on a board of frames and piles",
          "source_url": "https://hedwig-ai.com/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Agent that drafts replies and waits for approval before applying",
          "source_url": "https://hedwig-ai.com/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Connects through the official Google APIs; the FAQ says granted access is scoped to the features actually used.",
        "Sits on top of Gmail and Google Calendar without replacing them; accounts can be disconnected at any time with mail left as it was.",
        "Every draft waits for the user's explicit apply, and the vendor says the agent halts at anything it cannot undo."
      ],
      "limitations": [
        "Teams tier price and launch date unpublished (TBD); 'free forever for early birds' terms after the Teams launch are not specified.",
        "'Engineered to meet SOC 2 standards' is not evidence of a completed SOC 2 audit.",
        "Agent halting before irreversible actions is a vendor statement, not tested by AgentComparison.",
        "Described by the vendor; not tested by AgentComparison."
      ],
      "runtime": [
        "Web app on top of Gmail and Google Calendar per the vendor site"
      ],
      "privacy_security": "Hedwig's FAQ says user content is not used to train AI models: its AI providers are 'bound by data protection agreements that forbid using your data for their own purposes, including training', and content is never used for marketing or advertising. It says it is 'engineered to meet SOC 2 standards' with encryption in transit and at rest and least-privilege access - note this is an engineering claim, not a shown certification. Monetisation is stated as paid products, not data. Checked 22 October 2026.",
      "pricing": {
        "amount": 0,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Free plan: every feature, unlimited accounts and boards, agent included, no credit card",
        "usage_charges": "None listed",
        "verified_at": "2026-10-22",
        "source_url": "https://hedwig-ai.com/",
        "note": "hedwig-ai.com lists Free at $0 for early birds and a Teams tier coming soon with no price (TBD)."
      },
      "sources": [
        "https://hedwig-ai.com/"
      ],
      "changelog": [
        {
          "date": "2026-10-22",
          "change": "Field-level audit: live FAQ/pricing captured - Free $0 'forever, for early birds' (all features, agent included, no card), Teams TBD; Google-API scoped access, disconnect-any-time; explicit-apply drafts and halt-before-irreversible (vendor statements); no-training via provider DPAs, 'engineered to meet SOC 2' (not a shown certification). Permissions, coverage rows, setup steps, privacy fields filled. No hands-on testing."
        },
        {
          "date": "2026-10-04",
          "change": "Added at the owner's request after checking the vendor's own home page (pricing and FAQ sections). Documentation research only."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Free tier confirmed",
          "Live pricing section: Free $0 'forever, for early birds' - every feature, unlimited accounts and boards, agent included, no credit card or trial clock. Teams tier listed as 'coming soon', price TBD.",
          "https://hedwig-ai.com/"
        ],
        [
          "Agent autonomy",
          "Approval-gated drafts",
          "Vendor states every draft waits for explicit apply and the agent comes to a full stop before anything irreversible. This is a vendor statement, not an observed control.",
          "https://hedwig-ai.com/"
        ],
        [
          "Data use",
          "No training, per FAQ",
          "AI providers bound by data protection agreements forbidding use of user data for their own purposes including training; content never used for marketing or advertising.",
          "https://hedwig-ai.com/"
        ],
        [
          "Access model",
          "Google APIs, scoped",
          "Official Google APIs connect Gmail and Google Calendar; access scoped to features used; disconnect any time.",
          "https://hedwig-ai.com/"
        ]
      ],
      "setup_steps": [
        "Connect one Google account first and confirm the requested scopes match what you intend to use.",
        "Run one low-risk triage or draft task and check the draft waits for your explicit apply before anything sends.",
        "Review which frames are wired to agents before adding more accounts or boards."
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Free plan, $0 while new; Teams tier coming soon, no price (vendor pricing)",
      "page_url": "https://agentcomparison.net/agents/hedwig/"
    },
    {
      "id": "bolt",
      "name": "Bolt",
      "vendor": "StackBlitz",
      "kind": "product",
      "category": "AI app and website builder",
      "group": "coding-agents",
      "summary": "Bolt builds websites, apps and prototypes from a chat prompt, and can import from Figma or GitHub. The vendor site describes a Bolt Agent that routes between models, tests and iterates on its own output, and Bolt Cloud hosting with databases, authentication and integrations.",
      "one_liner": "Builds websites and apps from a chat prompt, with hosting and databases; priced by tokens.",
      "best_fit": "People who want a working web app or site from a prompt and are fine with token-based limits.",
      "poor_fit": "Anyone who needs predictable cost on large projects: token use grows with project size, per the vendor's own FAQ.",
      "decision_note": "Bolt prices by tokens: a free plan with 300K tokens a day and 1M a month (Bolt branding on sites), Pro at $18 a month billed annually ($216 a year, reduced from $300) with 10M tokens a month and rollover, and Teams at $27 per member per month billed annually (reduced from $30). Enterprise is by quote. Monthly-billing prices were not captured on this read.",
      "availability": "Web; paid plans and enterprise by quote (vendor pricing page).",
      "last_checked": "2026-10-22",
      "evidence_status": "hands_on_tested",
      "capabilities": [
        {
          "name": "Chat-to-app builder with Figma and GitHub import",
          "source_url": "https://bolt.new/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Token-based plans: Free (1M tokens/month), Pro and Teams billed annually",
          "source_url": "https://bolt.new/pricing",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Bolt Forge mode using open-weight AI models (eligible plans; enabling it consents to training/data-licensing use)",
          "source_url": "https://stackblitz.com/privacy-policy",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Eligible plans only",
          "region_restrictions": "Consent flow not presented to EEA/UK/Swiss accounts"
        }
      ],
      "permissions": [
        "Model-training and dataset-licensing use of your Bolt content can be turned off in account settings or via privacy@stackblitz.com (free, any plan); EEA/UK/Swiss accounts are excluded by default.",
        "Bolt Forge requires an explicit consent click ('Use Forge and consent to data use') before enabling; withdrawing consent stops future use but not processing already done.",
        "Enterprise tier documents SSO, audit logs and compliance support; granular per-action permission controls were not found in this check."
      ],
      "limitations": [
        "Free plan: 300K tokens daily and 1M monthly, Bolt branding on websites, 10MB file upload limit",
        "Token use rises with project size (vendor FAQ)",
        "Claims such as '98% less errors' and 'projects 1,000 times larger' are vendor marketing; not tested by AgentComparison",
        "Described by the vendor; not tested by AgentComparison",
        "Unless you opt out, Bolt content may train StackBlitz models and feed licensed de-identified datasets (EEA/UK/Swiss accounts excluded).",
        "bolt.new/privacy no longer resolves (policy lives at stackblitz.com/privacy-policy) and bolt.new/security now shows a database product page rather than trust documentation."
      ],
      "runtime": [
        "Web app with hosting, databases and import from Figma or GitHub per the vendor site"
      ],
      "privacy_security": "Read 22 October 2026 (StackBlitz privacy policy; bolt.new/privacy redirects away). Unless you opt out, Bolt prompts, code, outputs and interaction data ('Bolt Model Development Content', created on or after 7 October 2026) may be used to train StackBlitz's AI models and to build de-identified datasets licensed to third-party AI developers. Opting out is free on any plan via account settings or privacy@stackblitz.com, implemented within 15 days, but does not unwind models already trained or datasets already delivered. Accounts located in the EEA, UK or Switzerland are excluded from these uses entirely, as are organization-managed accounts (unless an admin enables them) and accounts with separate written agreements. Bolt Forge, an open-weight-model mode on eligible plans, requires affirmative consent to this data use as a condition of enabling it. Third-party AI providers process requests only to serve them and are contractually barred from training on them. Content is retained while the account is active, with a 30-day retrieval window after non-cause termination and residual backup copies for a limited period. Vendor policy, not independently audited.",
      "pricing": {
        "amount": 18,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Free 1M tokens a month (300K daily); Pro from 10M tokens a month with rollover; Teams 10M per member",
        "usage_charges": "Tokens; unused paid tokens roll over (vendor plan comparison)",
        "verified_at": "2026-10-22",
        "source_url": "https://bolt.new/pricing",
        "note": "bolt.new/pricing (read 22 October 2026, annual view): Free $0 (1M tokens/month, 300K daily, Bolt branding); Pro $18/month billed annually ($216/year, reduced from $300/year) with 10M tokens/month, rollover and branding removed; Teams $27/member/month billed annually ($324/year, reduced from $360/year); Enterprise custom. Monthly-billing prices sit behind a toggle and were not captured this read."
      },
      "sources": [
        "https://bolt.new/",
        "https://bolt.new/pricing",
        "https://stackblitz.com/privacy-policy",
        "https://bolt.new/terms"
      ],
      "changelog": [
        {
          "date": "2026-10-05",
          "change": "Added from the backlog after checking the vendor's own pages. Documentation research only."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: privacy terms established for the first time from the StackBlitz policy (Bolt content may train models and feed licensed de-identified datasets unless you opt out; EEA/UK/Swiss excluded; Forge consent gate; 15-day opt-out; retention terms). Added privacy/terms sources, Forge capability; noted bolt.new/privacy and /security URL rot."
        },
        {
          "date": "2026-10-24",
          "note": "Hands-on pilot added: three build-and-edit runs on the free plan passed the fixed price/element/edit checks (3/3); unbriefed invented business claims recorded separately; cost observations on the free allowance. Evidence status now hands-on pilot completed. Pilot, not a full evaluation; no overall score."
        },
        {
          "date": "2026-10-24",
          "note": "Hands-on wording narrowed on review: run 3's form check was page-observation only (network instrumentation from runs 1-2 not repeated); accordion expand/collapse not separately clicked. Verified results unchanged."
        },
        {
          "date": "2026-10-24",
          "note": "Hands-on table: form row heading/result now separate the two evidence levels (network checked in 2 runs; page feedback in all 3) instead of a single 3/3 verdict."
        }
      ],
      "price_tags": [
        "free",
        "listed",
        "usage"
      ],
      "price_summary": "Free (1M tokens/month); Pro $18/month billed annually; Teams $27/member/month billed annually (vendor pricing)",
      "hands_on": {
        "tested_label": "Hands-on pilot, 22-24 October 2026",
        "title": "Hands-on pilot: three build-and-edit runs on the free plan",
        "intro": "We ran three independent build-and-edit pairs on Bolt's free plan (one fresh project each, 22-24 October 2026) with a fixed fictional brief: a one-page site for a Bristol dog-walking business with three named prices, a contact form and a footer phone number, followed by a fixed edit (change one price, add a two-question FAQ, leave the other prices alone). Every result was checked in the rendered page itself - structure and screenshots - never from the tool's own summary.",
        "checks": [
          [
            "Exact prices from the brief (£12, £18, £35) shown with the right service",
            "Pass, 3 of 3 runs",
            "Each price sat inside its own labelled card in every run."
          ],
          [
            "Required elements: headline, three-field contact form, footer phone number",
            "Pass, 3 of 3 runs",
            "Form fields were exactly name, email and message each time."
          ],
          [
            "Edit precision: Boarding changed to £40, the other two prices untouched",
            "Pass, 3 of 3 runs",
            "Walk and Adventure stayed at £12 and £18 in every run."
          ],
          [
            "New section on request: FAQ with exactly two questions",
            "Pass, 3 of 3 runs",
            "Two questions present in each run, rendered accordion-style (first item open by default where observed); expand/collapse interaction was not separately clicked. The edit also added a navigation link."
          ],
          [
            "Demo-form submission behaviour",
            "Network checked in 2 runs; page feedback checked in all 3",
            "Runs 1 and 2: instrumented submit test logged zero network calls. Run 3: checked in the page only - the confirmation itself stated nothing was sent - without repeating the network instrumentation. What a user sees after submitting varied: one silent run, one 'Message sent!' with no note that nothing was sent, one clearly labelled demo form."
          ]
        ],
        "observations": [
          "Every run invented business claims we did not ask for: DBS checks, insurance, star ratings, '500+ walks', and a made-up contact email and postcode. These sit outside the brief-following checks, but you must review any business text Bolt writes before publishing it.",
          "Cost: the free plan covered all three pairs. Each pair used roughly 155k-195k tokens of the 300k daily free allowance (1M monthly). Figures are approximate, read from Bolt's rounded balance display."
        ],
        "limits": "A pilot, not a verdict: one product, one brief, three runs, free plan. No overall score and no ranking. Bolt's training opt-out exists in settings and was verified off before the later runs; our first run preceded that check, so we cannot prove the switch's state for it.",
        "faq_answer": "Yes - a small hands-on pilot on the free plan (22-24 October 2026). Bolt passed our fixed layout and price-edit checks in all three runs, and every run also invented business claims we did not ask for. Details in the hands-on section above. A three-run pilot is not a full evaluation."
      },
      "page_url": "https://agentcomparison.net/agents/bolt/"
    },
    {
      "id": "perplexity-comet",
      "name": "Perplexity Comet",
      "vendor": "Perplexity",
      "kind": "product",
      "category": "AI browser with an assistant",
      "group": "errands-travel-food",
      "summary": "Comet is Perplexity's browser, available for Mac, Windows, iOS and Android. The vendor page shows an assistant that answers questions about what you are reading, drafts email replies, builds simple sites, makes study plans and shops, and Perplexity's pricing page lists a Comet Agent and Comet Assistant among plan features.",
      "one_liner": "Perplexity's browser with an assistant that reads, drafts and shops for you; free to download.",
      "best_fit": "People who want an AI assistant inside the browser to read, summarise, draft and shop, and who are comfortable giving a browser assistant access to their sessions.",
      "poor_fit": "Anyone who needs the exact free-versus-paid Comet limits: the vendor pages we checked do not spell out which agent actions the free tier allows.",
      "decision_note": "Perplexity says the Comet browser is free to download for everyone (its blog post is dated 2 Oct 2025). Its pricing page lists Free at $0, Pro at $20 a month and Max at $200 a month, with the stronger models behind Comet Assistant and Comet Agent listed under the paid plans; Max adds a 'Max Assistant' mode on Comet per the vendor help centre.",
      "availability": "Free download for Mac, Windows, iOS and Android; paid Perplexity plans add stronger assistant models (vendor pages).",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "AI assistant inside a desktop and mobile browser",
          "source_url": "https://www.perplexity.ai/comet",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Free browser; stronger models on paid plans",
          "source_url": "https://www.perplexity.ai/hub/pricing",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Max Assistant mode on Comet",
          "source_url": "https://www.perplexity.ai/help-center/en/articles/11680686-perplexity-max.html",
          "observed_at": "2026-10-05",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Max plan",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Comet has privacy settings (Settings, Privacy and Security) including third-party cookie controls, plus Incognito mode in the browser, app and parts of the website; targeted-advertising opt-outs are honoured via cookie controls and Global Privacy Control.",
        "Email Assistant is opt-in per linked email provider; its email content is excluded from AI training per the privacy notice.",
        "Granular controls over what the assistant may do with page content or saved logins were not established in this check."
      ],
      "limitations": [
        "Which Comet Agent actions are available on the free plan is not stated on the pages we checked",
        "The pricing page we read is in Spanish; figures are the vendor's ($0, $20, $200 a month)",
        "An assistant that acts in your browser can reach pages you are signed in to; no permission list was found in this check",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Desktop and mobile browser per the vendor page"
      ],
      "privacy_security": "Perplexity Privacy Notice (updated 19 October 2026, read 22 October 2026; covers Comet but not Enterprise/API). If you link Email Assistant, your email content is not used to create, train, improve or fine-tune AI models, and Google account data is handled under Google's Limited Use requirements. Comet processes saved settings, passwords, payment and account information you store in the browser; depending on your settings it records browsing history and engagement when used with a synced account. Some data (browsing data, site permissions, download records) is stored locally on your device; the notice says Perplexity does not access it, while also saying such interaction data may be used to provide and improve Comet - the two sentences sit uneasily together. Generally, Perplexity uses data to 'improve or create services and products, including our AI models'. Retention is as long as needed for the stated purposes; deletion can be requested and is granted where required by law, with business/legal exceptions. Vendor policy, not independently audited.",
      "pricing": {
        "amount": 0,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Comet browser free; Perplexity Free $0, Pro $20 a month, Max $200 a month",
        "usage_charges": "Not stated for Comet specifically",
        "verified_at": "2026-10-22",
        "source_url": "https://www.perplexity.ai/comet",
        "note": "perplexity.ai/comet (read 22 October 2026) lists Free at $0, Select at $17/month billed annually (expanded Computer access, 4,000 bonus credits, top models) and Max at $167/month billed annually (maximum Computer usage, 35,000 bonus credits). The earlier Pro $20/Max $200 naming is no longer shown; monthly-billing prices were not captured. The Comet browser remains free to download."
      },
      "sources": [
        "https://www.perplexity.ai/comet",
        "https://www.perplexity.ai/hub/blog/comet-is-now-available-to-everyone-worldwide",
        "https://www.perplexity.ai/hub/pricing",
        "https://www.perplexity.ai/help-center/en/articles/11680686-perplexity-max.html",
        "https://www.perplexity.ai/hub/legal/privacy-notice"
      ],
      "changelog": [
        {
          "date": "2026-10-05",
          "change": "Added from the backlog after checking the vendor's own pages. Documentation research only."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: privacy notice read for the first time (updated 19 October 2026) - Email Assistant content excluded from AI training; Comet saved-info processing, synced browsing-history recording and local-data wording documented; privacy settings and Incognito options added to permissions."
        }
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Browser free; Select $17/month and Max $167/month, both billed annually (vendor page)",
      "page_url": "https://agentcomparison.net/agents/perplexity-comet/"
    },
    {
      "id": "halo",
      "name": "Halo",
      "vendor": "Halo",
      "kind": "product",
      "category": "on-device personal assistant for iPhone",
      "group": "personal-assistants",
      "summary": "Halo is an iPhone app that reads mail, calendar and reminders, writes a daily briefing and runs automations you describe in plain words. The vendor says its data stays on the device and that you bring your own AI provider (ChatGPT, Grok or Apple Intelligence); it also connects to other apps through the Model Context Protocol.",
      "one_liner": "An iPhone assistant that writes a daily briefing and runs automations, keeping data on the phone.",
      "best_fit": "iPhone users who want a daily briefing across mail, calendar and reminders and who prefer an assistant that keeps data on the phone and uses their own AI account.",
      "poor_fit": "Anyone on Android or an older iPhone, or who wants a single price with the AI model included: you bring and pay for your own AI provider.",
      "decision_note": "The vendor's site lists a one-time price of $49 (early bird, shown against $99) after a 7-day free trial, with no subscription, while the App Store listing shows a free download with a 'Halo Pro' in-app purchase at $49.99. You also supply your own AI provider access (ChatGPT, Grok or Apple Intelligence), so any cost your provider charges is separate from Halo's price. The vendor says it runs no servers; AI requests still go from your phone to the provider you pick. Halo is not in our text-message assistants table: its documented interface is an iPhone app that produces a daily briefing and runs described automations; no two-way messaging channel is documented.",
      "availability": "iPhone only (App Store listing requires iOS 26.0 or later); 7-day trial then a one-time purchase per the vendor site.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Daily briefing across mail, calendar and reminders",
          "source_url": "https://meethalo.app/",
          "observed_at": "2026-10-05",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Automations described in plain words, using the Model Context Protocol to reach other apps",
          "source_url": "https://meethalo.app/",
          "observed_at": "2026-10-05",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Bring-your-own AI provider: ChatGPT, Grok or Apple Intelligence",
          "source_url": "https://meethalo.app/",
          "observed_at": "2026-10-05",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Reads on-phone sources (mail, calendar, reminders; health, music, notes marked coming soon) and reaches apps via Model Context Protocol (Notion, Slack, Spotify, Uber, Swiggy, Zomato listed).",
        "Writes a single daily briefing and speaks up only when something needs you.",
        "Plain-words automations that fire on triggers (e.g. digest spending at 9pm, sync transactions to Notion at 1am, prep tomorrow at 10pm)."
      ],
      "limitations": [
        "iPhone only; the App Store listing requires iOS 26.0 or later",
        "Health, Music and Notes sources are marked 'coming soon' on the vendor site",
        "You supply your own AI provider; requests go to that provider under its own privacy policy",
        "Vendor site and App Store show slightly different prices ($49 one-time versus $49.99 Halo Pro)",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "iPhone app that runs on the device, with no Halo servers per the vendor privacy policy",
        "Connects to other apps through the Model Context Protocol (MCP)"
      ],
      "privacy_security": "Halo's privacy policy (last updated 26 June 2026) is local-first: no accounts, no server storage; emails, calendar, reminders, health data and memories stay on the iPhone and sync only through the user's private iCloud (a user-Apple relationship Halo cannot access). Connections are made on-device and AI requests go directly from the device to the provider the user configures (OpenAI, xAI, Apple Intelligence) with the user's own key or subscription, so Halo never sees request contents; the provider's own policy governs that traffic. Only anonymous crash reports and aggregate usage analytics are collected, never private content or API keys. Security page: 'we don't store your data, so there's nothing to breach.' Checked 23 October 2026.",
      "pricing": {
        "amount": 49,
        "currency": "USD",
        "billing_period": "one-time",
        "allowances": "7-day free trial with everything unlocked, then a one-time purchase",
        "usage_charges": "Your own AI provider's costs are separate",
        "verified_at": "2026-10-23",
        "source_url": "https://meethalo.app/",
        "note": "Free download on the App Store (v1.2.1, checked 23 October 2026) with in-app purchases; a $6.99 IAP is listed. meethalo.app/pricing was Mod_Security-blocked from automated fetches, so exact tiers are unverified. AI usage runs through the user's own ChatGPT/Grok/Apple Intelligence subscription."
      },
      "sources": [
        "https://meethalo.app/",
        "https://meethalo.app/privacy/",
        "https://meethalo.app/security/",
        "https://apps.apple.com/app/id6783715054"
      ],
      "changelog": [
        {
          "date": "2026-10-05",
          "change": "Added after the developer announced App Store availability on 5 October 2026 (used as a lead only), then checking the vendor's own site and App Store listing. Documentation research only."
        }
      ],
      "coverage_rows": [
        [
          "Architecture",
          "Local-first, no accounts",
          "No account or signup; data stays on the iPhone and syncs via the user's own private iCloud (between user and Apple). Halo never has server access to user data.",
          "https://meethalo.app/privacy/"
        ],
        [
          "AI model",
          "Bring your own",
          "Requests go directly from the device to the provider you configure - ChatGPT/OpenAI, Grok/xAI or Apple Intelligence - using your own key or subscription; Halo never sees request contents.",
          "https://meethalo.app/"
        ],
        [
          "Analytics",
          "Anonymous only",
          "Only anonymous crash reports and aggregate usage info; never emails, messages, calendar events, health data or API keys.",
          "https://meethalo.app/privacy/"
        ],
        [
          "Pricing",
          "Free + IAP",
          "Free download on the App Store (v1.2.1) with in-app purchases; a $6.99 IAP is listed. The vendor pricing page was bot-blocked (Mod_Security), so exact tiers are unverified. AI usage is billed through your own provider subscription.",
          "https://apps.apple.com/app/id6783715054"
        ],
        [
          "Memory",
          "On-device wiki",
          "Structured human-readable Markdown memory of people, health, work, finances and travel, stored on-device.",
          "https://meethalo.app/"
        ]
      ],
      "setup_steps": [
        "Download Halo from the App Store (free, in-app purchases; no account or signup).",
        "Plug in your own AI: ChatGPT/OpenAI, Grok/xAI or Apple Intelligence - requests go straight from your phone to that provider.",
        "Connect on-phone sources (mail, calendar, reminders) and MCP apps like Notion, Slack, Spotify or Uber.",
        "Describe automations in plain words; review the on-device Markdown memory anytime."
      ],
      "price_tags": [
        "listed"
      ],
      "price_summary": "7-day free trial, then $49 one-time (vendor site); you also pay your own AI provider",
      "page_url": "https://agentcomparison.net/agents/halo/"
    },
    {
      "id": "hark",
      "name": "Hark",
      "vendor": "Hark Labs, Inc.",
      "kind": "product",
      "category": "personal AI with a browser-using agent",
      "group": "personal-assistants",
      "summary": "Hark is a personal AI from Hark Labs, founded by Figure AI's Brett Adcock, that is not yet generally available. What Hark has published is a research preview of Handoff, a computer use agent that runs each task on its own virtual computer with a browser, file system and terminal and can use your connected accounts; the vendor lists ordering food, shopping, restaurant bookings, travel and research as examples.",
      "one_liner": "A well-funded personal AI with a browser-using agent for web tasks; waitlist, no published price yet.",
      "best_fit": "People who want to follow a heavily funded personal assistant that works through a browser, and who are comfortable waiting for a published price, permissions list and independent testing before trusting it with accounts.",
      "poor_fit": "Anyone who needs a product they can use and price today: no plan, price or detailed approval controls are published yet.",
      "decision_note": "As of 5 October 2026 Hark's own site still shows only a sign-up form, and no price or plan list is published on the pages we read. The founder said on X on 4 October that Hark launches this week and that the first 100,000 sign-ups get a paid plan free; the plan length and features were not stated, per RuntimeWire's report of the post. The Handoff preview, its benchmark claims and the demos are vendor material; TechCrunch noted the demo videos show only part of the process. Hark is not in our text-message assistants table: it is waitlist-only with a research preview of a browser-using agent, and no two-way messaging channel is documented.",
      "availability": "Waitlist; the founder says launch is this week (X post, 4 October 2026). Handoff was shown as a research preview on 5 August 2026.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Computer use agent (Handoff) that completes web tasks on a dedicated virtual computer",
          "source_url": "https://hark.com/articles/introducing-hark-handoff",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Connects your existing accounts so tasks use saved addresses, preferences and history",
          "source_url": "https://hark.com/articles/introducing-hark-handoff",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Example tasks listed by the vendor: food orders, shopping, restaurant reservations, travel, research and LinkedIn outreach",
          "source_url": "https://hark.com/articles/introducing-hark-handoff",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Handoff spins up a dedicated virtual computer with its own browser, file system and terminal, and can log into your existing accounts on your behalf.",
        "Terms confirm the service may store third-party login credentials, session cookies, authentication tokens and payment card details ('Stored Credentials') to act on your behalf; removable via account settings.",
        "Vendor states Hark checks in with you before doing something that requires your approval; Terms confirm the agent can execute transactions using your saved payment method when you instruct it.",
        "Homepage shows integrations with services such as Strava, Spotify and Venmo via Panels."
      ],
      "limitations": [
        "No public sign-up or pricing on the site (Terms describe subscriptions and trial-to-paid conversion, so access is presumably gated).",
        "End-to-end encryption, approval check-ins and benchmark results are vendor statements, not tested by AgentComparison.",
        "Agent transactions use your saved payment method when instructed - review the Terms' transaction and merchant-disclaimer sections before enabling payments.",
        "Described by the vendor; not tested by AgentComparison."
      ],
      "runtime": [
        "Each Handoff request runs on a dedicated virtual computer with its own browser, file system and terminal, per Hark's research-preview post",
        "Hark's terms also describe optional local-access features that use your existing logins on your device"
      ],
      "privacy_security": "'Secured by hark' is described on the homepage as end-to-end encrypted - 'a vault that absolutely no person has access to' - covering connected accounts, saved logins and payments. The Terms confirm the service may store login credentials, session cookies, authentication tokens and payment card details for third-party sites ('Stored Credentials'), protected by reasonable safeguards, removable at any time via account settings and deleted on removal or account closure. Hark also states it checks in before anything requiring approval. These are vendor statements, not independently verified. Checked 22 October 2026.",
      "pricing": {
        "amount": null,
        "currency": "USD",
        "billing_period": null,
        "allowances": null,
        "usage_charges": "Not stated",
        "verified_at": "2026-10-05",
        "source_url": "https://runtimewire.com/article/hark-first-100000-signups-free-paid-plan",
        "note": "No price is published on hark.com (checked 22 October 2026). RuntimeWire reported the founder's 4 October post that the first 100,000 sign-ups get a paid plan free, without a duration or feature list; that article URL now redirects to hark.com, so the report can no longer be re-read at its original address."
      },
      "sources": [
        "https://hark.com/",
        "https://hark.com/articles/introducing-hark-handoff",
        "https://hark.com/terms",
        "https://hark.com/articles/hark-announces-usd700-million-fundraising-round",
        "https://techcrunch.com/2026/08/05/hark-previews-its-browser-use-agent-for-completing-tasks/",
        "https://runtimewire.com/article/hark-first-100000-signups-free-paid-plan"
      ],
      "changelog": [
        {
          "date": "2026-10-22",
          "change": "Field-level audit: homepage + Handoff article + Terms reviewed. Still no public sign-up/pricing (Terms confirm subscription/trial model). Handoff runs a dedicated virtual computer and can log into your accounts; Stored Credentials (logins, session cookies, auth tokens, card details) held with reasonable safeguards and removable; claimed E2E 'Secured by hark' vault; approval check-ins claimed. All vendor statements, no hands-on testing."
        },
        {
          "date": "2026-10-05",
          "change": "Added from Hark's own pages and terms plus press coverage (TechCrunch, RuntimeWire, TestingCatalog used for context). Documentation research only; the product is not yet generally available."
        }
      ],
      "coverage_rows": [
        [
          "Sign-up & pricing",
          "Still not publicly shown",
          "No public sign-up or pricing on the homepage as of 22 October 2026; the Terms describe a subscription with trial-to-paid conversion, so the commercial model exists but pricing is not public.",
          "https://hark.com/"
        ],
        [
          "Browser automation",
          "Handoff is the browser agent",
          "Hark's handoff article: each run spins a dedicated virtual computer (browser, file system, terminal); vendor-reported Online-Mind2Web top spot, ahead of GPT 5.4 by 8 points per its own post.",
          "https://hark.com/articles/introducing-hark-handoff"
        ],
        [
          "Credentials handling",
          "Stored Credentials + claimed E2E vault",
          "Terms allow storing third-party credentials/tokens/card details with reasonable safeguards and deletion on removal; homepage claims end-to-end encryption no person can access. Both vendor statements.",
          "https://hark.com/terms"
        ],
        [
          "Action approval",
          "Vendor claims check-ins",
          "Homepage: 'Hark checks in with you before doing something that requires your approval.' Not tested by AgentComparison.",
          "https://hark.com/"
        ]
      ],
      "setup_steps": [
        "Read the Terms' Stored Credentials section before connecting any account - saved logins, session cookies and card details may be retained until you remove them.",
        "Start with a low-stakes Handoff task that touches no payment method, and watch how the approval check-ins behave.",
        "Remove any Stored Credentials you no longer want held via account settings when finished."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Price not published; founder says first 100,000 sign-ups get a paid plan free (length not stated)",
      "meta_title": "Hark: browser-using personal AI agent, waitlist status",
      "page_url": "https://agentcomparison.net/agents/hark/"
    },
    {
      "id": "photon",
      "name": "Photon (Spectrum)",
      "vendor": "Photon",
      "kind": "product",
      "category": "messaging infrastructure for AI agents",
      "group": "builders",
      "summary": "Photon provides an open-source framework called Spectrum, plus hosted iMessage lines, that lets a developer connect one agent to iMessage, SMS/RCS, WhatsApp Business, Telegram and other chat interfaces. It is infrastructure for people building agents, not an assistant you use directly.",
      "one_liner": "Open-source framework and hosted iMessage lines for connecting your own agent to chat apps.",
      "best_fit": "Developers who want one agent codebase reachable over iMessage and other chat apps without building each integration, and who are comfortable with a hosted messaging dependency or self-hosting.",
      "poor_fit": "Anyone looking for a finished assistant to use, or teams that need independently audited uptime, deliverability and data-handling terms before routing customer conversations through a third party.",
      "decision_note": "Developer infrastructure, not a ready-made assistant. Photon's pricing page lists a free tier, Pro at $25 a month and Business at $250 per dedicated line per month, with Enterprise on request. The Spectrum core is MIT-licensed on GitHub and can be self-hosted. What you cannot judge from the vendor's pages is message deliverability and number reputation at scale, so read its iMessage deliverability docs and test on your own traffic before building a product on it.",
      "availability": "Hosted dashboard and API at photon.codes; the Spectrum core is open source and can be self-hosted. Photon's docs say Spectrum supports iMessage, WhatsApp Business, Telegram, a terminal provider and SIP voice calls on iMessage lines.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "One agent connected to iMessage, WhatsApp Business, Telegram and a terminal provider through Spectrum",
          "source_url": "https://photon.codes/docs",
          "observed_at": "2026-10-05",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "All plans; Business adds WhatsApp and phone call options",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Open-source Spectrum core under the MIT licence, self-hostable",
          "source_url": "https://github.com/photon-hq/spectrum-ts",
          "observed_at": "2026-10-05",
          "evidence_type": "repository_metadata",
          "confidence": "high",
          "plan_restrictions": "None for self-hosting",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Managed iMessage lines with SMS/RCS fallback, polls and other iOS 26 features",
          "source_url": "https://photon.codes/platform/imessage",
          "observed_at": "2026-10-05",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Free, Pro and Business tiers differ by users and number type",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Dedicated iMessage lines you own, with group messaging and limited cold outreach",
          "source_url": "https://photon.codes/pricing",
          "observed_at": "2026-10-05",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Business ($250 per line per month) and Enterprise",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Gives a developer's agent hosted iMessage lines with RCS/SMS fallback and full direct-messaging API access.",
        "Business tier supports cold outreach to up to 50 new contacts per day and full group messaging.",
        "Bring-your-own iMessage Mini Apps on the approved App Hub."
      ],
      "limitations": [
        "Infrastructure for builders: you write and run the agent yourself and pay model costs separately",
        "The 99.95% uptime figure on photon.codes is the vendor's own claim; we did not find an independent status history",
        "Free tier uses shared managed numbers and is capped at 10 users, Pro at 100 users; unlimited users need the Business plan with Auto Scale, per the pricing page",
        "Cold outreach is limited to 50 new contacts per day per line on Business, per the pricing page",
        "Open-source self-hosting runs on your own Mac and uses your iCloud account as the iMessage number, per the pricing page; Photon's docs separate this from its managed lines",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Hosted Photon dashboard and API, or self-hosted Spectrum (TypeScript, MIT licence on GitHub)",
        "Providers for iMessage, WhatsApp Business, Telegram, terminal and SIP voice, plus custom providers"
      ],
      "privacy_security": "No privacy policy was found on photon.codes - /privacy 404s as of 23 October 2026. Photon is developer infrastructure (hosted iMessage lines plus the open-source Spectrum framework), so message and contact data handling is undocumented on the public site; treat retention, training use and sharing as unknown. Security-relevant facts from the platform pages: dedicated lines on Business, cold outreach capped at 50 new contacts/day, Auto Scale required for unlimited users.",
      "pricing": {
        "amount": 25,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Pro: up to 100 users, managed shared numbers",
        "usage_charges": "Business is $250 per line per month; Enterprise is custom. Your own model costs are separate.",
        "verified_at": "2026-10-23",
        "source_url": "https://photon.codes/pricing",
        "note": "photon.codes/pricing (checked 23 October 2026): Free $0 (iMessage unlimited daily with Auto Scale, RCS/SMS fallback, up to 10 users); Pro $25/mo (up to 100 users); Business $250/line/mo (dedicated iMessage lines, group messaging API, cold outreach up to 50 new contacts/day, unlimited users with Auto Scale); Enterprise custom."
      },
      "sources": [
        "https://photon.codes/",
        "https://photon.codes/docs",
        "https://photon.codes/platform/imessage",
        "https://photon.codes/pricing",
        "https://github.com/photon-hq/spectrum-ts"
      ],
      "changelog": [
        {
          "date": "2026-10-05",
          "change": "Added from Photon's own site, docs and pricing page and the public GitHub repository. Documentation research only; not tested."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Free / $25 / $250",
          "Free $0 (iMessage unlimited daily messages with Auto Scale, RCS/SMS fallback, up to 10 users); Pro $25/mo (up to 100 users); Business $250/line/mo (dedicated iMessage lines, group messaging, cold outreach up to 50 new contacts/day, unlimited users with Auto Scale); Enterprise custom.",
          "https://photon.codes/pricing"
        ],
        [
          "Model",
          "Developer platform",
          "Hosted iMessage lines plus the open-source Spectrum framework connect one agent to iMessage, SMS/RCS, WhatsApp, Telegram, Slack and Discord - infrastructure, not an end-user assistant.",
          "https://photon.codes/"
        ],
        [
          "Mini Apps",
          "Approved iMessage App Hub",
          "Photon runs an App Hub for iMessage, officially approved on the iMessage App Store, so developers can share Mini Apps.",
          "https://photon.codes/platform/imessage"
        ],
        [
          "Privacy policy",
          "Not found",
          "photon.codes/privacy 404s as of 23 October 2026; no privacy policy was located on the site. Data-handling practices are undocumented - treat as unknown.",
          "https://photon.codes/"
        ]
      ],
      "setup_steps": [
        "Start free: iMessage with unlimited daily messages (Auto Scale), RCS/SMS fallback, up to 10 users.",
        "Pro $25/mo lifts to 100 users; Business $250/line/mo adds dedicated lines, group messaging and cold outreach (50/day).",
        "Docs and the open-source Spectrum framework (github.com/photon-hq/spectrum-ts) cover integration."
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Free tier; Pro $25/month; Business $250 per line per month (vendor pricing page)",
      "meta_title": "Photon (Spectrum): your agent on iMessage and WhatsApp",
      "page_url": "https://agentcomparison.net/agents/photon/"
    },
    {
      "id": "open-instinct",
      "name": "Open Instinct (Maria Gorskikh)",
      "vendor": "Maria Gorskikh (open-source project)",
      "kind": "product",
      "category": "open-source personal agent you text",
      "group": "personal-assistants",
      "summary": "Open Instinct is an MIT-licensed, self-hosted personal agent you text on iMessage. Its README says each agent has its own computer (a microVM with a Linux desktop), connects to your apps through Composio, can pay with one-time Stripe Link cards you approve, and coordinates with the agents of people you trust. It describes itself as a from-scratch clone of Instinct, and says it is not affiliated with Instinct or with Merit Systems' separate OpenInstinct project.",
      "one_liner": "A free, self-hosted, MIT-licensed personal agent you text on iMessage; version 0.1.",
      "best_fit": "Developers and technically confident users who want to run and customise their own text-message personal agent, read its prompt and permission code, and control the keys.",
      "poor_fit": "Anyone who wants a hosted assistant that works after sign-up, or who is not ready to manage API keys, phone-line provisioning and a hosting account.",
      "decision_note": "A new open-source project, created on 3 October 2026 and labelled version 0.1 by its README, which says live iMessage and Maritime deployment are beta. The code is free (MIT) but you run it: you bring your own model key and need paid or metered accounts for the messaging line and hosting. It is not the hosted Instinct assistant and is not operated by Instinct. Read the permissions and security docs and start with the laptop mode before giving it an iMessage line or a wallet.",
      "availability": "Open-source code on GitHub (MIT). Self-hosted: on a laptop with a model key only, with an iMessage line through Inkbox, or one agent per person on Maritime. No hosted service from the project was found.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Personal agent you text over iMessage, with its own computer",
          "source_url": "https://github.com/mariagorskikh/open-instinct",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Free software; you pay for the services it runs on",
          "region_restrictions": "Not stated"
        },
        {
          "name": "MIT-licensed open-source code that anyone can fork",
          "source_url": "https://github.com/mariagorskikh/open-instinct",
          "observed_at": "2026-10-23",
          "evidence_type": "repository_metadata",
          "confidence": "high",
          "plan_restrictions": "None",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Trust tiers and plain-English grants for other people's agents",
          "source_url": "https://github.com/mariagorskikh/open-instinct/blob/main/README.md",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Free software; you pay for the services it runs on",
          "region_restrictions": "Not stated"
        },
        {
          "name": "One-time Stripe Link card payments the owner approves each time",
          "source_url": "https://github.com/mariagorskikh/open-instinct/blob/main/docs/PAYMENTS.md",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "US and Canada consumers for Link Agent Wallet",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "You text a phone number; Inkbox assigns the agent its number, email address and agent-to-agent endpoint.",
        "Runs in its own microVM (Maritime) with a Linux desktop; apps connect by name through Composio ('connect my Notion').",
        "Seven permission tiers enforced in code: your own agents act as staff, a partner's agent can read your calendar, a friend's can only ask when you are free, strangers get a polite no and you get notified who asked for what.",
        "Payments: the agent never holds your card - Stripe Link single-use card per purchase, one-tap approval, dead after; without a wallet it hands you the checkout screen instead.",
        "Can make phone calls via Inkbox Voice AI (e.g. call a restaurant to book) and texts back the outcome."
      ],
      "limitations": [
        "Version 0.1, created on 3 October 2026; the README calls live iMessage and Maritime deployment beta",
        "Self-hosted: you supply and pay for the model, Inkbox messaging line, Maritime hosting and any payment setup",
        "The 837-test and end-to-end claims in the README are the author's own; we did not run them",
        "Depends on several third-party services (Inkbox, Maritime, Composio, Stripe Link, a model provider)",
        "Stripe Link Agent Wallet is available to consumers in the United States and Canada, per the project's payments doc",
        "Not tested by AgentComparison"
      ],
      "runtime": [
        "Self-hosted TypeScript agent (Node 22.19+, pnpm 10) built on the Pi agent loop",
        "Agent runs in a Maritime microVM with a Linux desktop, or locally; iMessage, SMS, email and agent-to-agent transport come from Inkbox; apps through Composio"
      ],
      "privacy_security": "Open-source (MIT) and self-hosted: you bring your own keys and no keys are stored in the repository. The agent runs in its own microVM on Maritime with a Linux desktop, with a policy guard in front of every tool. A seven-tier permission model (staff/partner/friend/stranger plus plain-English grants) is enforced in code before any tool runs, and the owner gets a one-line text naming who asked for what. Payments never touch a stored card: Stripe Link issues a single-use card for the exact total, approved with one tap and dead after payment; spend policy file sets ask-above amounts, daily caps and always-ask items. Checked 23 October 2026.",
      "pricing": {
        "amount": 0,
        "currency": "USD",
        "billing_period": null,
        "allowances": null,
        "usage_charges": "The software is free under the MIT licence. The project's keys doc says the model is the biggest cost, Inkbox has a paid plan for iMessage lines and Maritime bills by the hour the VM is awake; no total is given.",
        "verified_at": "2026-10-05",
        "source_url": "https://github.com/mariagorskikh/open-instinct/blob/main/docs/KEYS.md",
        "note": "No price for the agent itself. Running it costs whatever your model, messaging line and hosting accounts charge; the keys doc gives no fixed total.",
        "class": "see_vendor"
      },
      "sources": [
        "https://github.com/mariagorskikh/open-instinct",
        "https://github.com/mariagorskikh/open-instinct/blob/main/README.md",
        "https://github.com/mariagorskikh/open-instinct/blob/main/docs/KEYS.md",
        "https://github.com/mariagorskikh/open-instinct/blob/main/docs/PAYMENTS.md",
        "https://x.com/mariagorskikh/status/2107116096990175454"
      ],
      "changelog": [
        {
          "date": "2026-10-23",
          "change": "Field-level audit from GitHub README/docs: MIT open-source Instinct-like personal agent over iMessage; Inkbox number/email/agent endpoint; Maritime microVM + Pi loop + Composio apps; seven-tier permission model enforced in code with owner notifications; Stripe Link single-use-card payments with one-tap approval and spend policy; Inkbox Voice AI phone calls; BYO keys, no keys in repo; default model Claude Fable 5.1. Coverage rows, permissions, setup steps, privacy fields filled. No hands-on testing."
        },
        {
          "date": "2026-10-05",
          "change": "Added from the project's GitHub repository, README and docs, plus the maintainer's X post. Documentation research only; not tested."
        }
      ],
      "coverage_rows": [
        [
          "Cost",
          "Free/open-source, BYO keys",
          "MIT licensed, self-hosted; you bring your own API keys (Anthropic default - Claude Fable 5.1, any Pi provider works). Stripe Link OAuth client requires a Stripe application reviewed in about 3-5 days.",
          "https://github.com/mariagorskikh/open-instinct"
        ],
        [
          "Permission model",
          "Seven tiers in code",
          "staff/partner/friend/stranger tiers enforced before any tool runs, plus plain-English grants ('Sam can book us dinner this week'); owner notified of every outside request.",
          "https://github.com/mariagorskikh/open-instinct"
        ],
        [
          "Payments",
          "Single-use cards",
          "Stripe Link wallet issues a single-use card for the exact total per purchase; one-tap approval; spend policy file for caps and always-ask items.",
          "https://github.com/mariagorskikh/open-instinct"
        ],
        [
          "Channels",
          "iMessage + voice calls",
          "Text via its Inkbox-provided number; can place real phone calls (restaurant bookings) and texts back outcome and follow-ups.",
          "https://github.com/mariagorskikh/open-instinct"
        ]
      ],
      "setup_steps": [
        "Clone the repo, install with Node 22.19+ and pnpm 10, and bring your own keys (KEYS.md lists them); try it locally with `pnpm instinct dev` before connecting a phone line.",
        "Apply for the Stripe Link OAuth client early (3-5 day review) if you want payments; without it the agent still reaches checkout and hands you the screen.",
        "Set your spend policy file and permission tiers before giving partner or friend agents access - defaults are restrictive, grants are explicit."
      ],
      "price_tags": [
        "free",
        "usage"
      ],
      "price_summary": "Free MIT software; you pay for your own model, messaging line and hosting",
      "page_url": "https://agentcomparison.net/agents/open-instinct/"
    },
    {
      "id": "openinstinct",
      "name": "OpenInstinct (Merit Systems)",
      "vendor": "Merit Systems (open-source project)",
      "kind": "product",
      "category": "open-source iMessage assistant with a browser and password vault",
      "group": "personal-assistants",
      "summary": "OpenInstinct is an MIT-licensed personal assistant that lives in an iMessage thread and uses a cloud browser to do chores such as booking tickets, ordering groceries and filling forms. You deploy it into your own Vercel account; passwords and cards go into an encrypted vault, and purchases, email, calendar events and destructive changes wait for your approval in the thread, per the project's site. It is not the hosted Instinct assistant and is separate from Maria Gorskikh's Open Instinct project.",
      "one_liner": "A free, self-hosted iMessage assistant that uses a cloud browser and an encrypted vault; needs Vercel Pro.",
      "best_fit": "People comfortable deploying to their own Vercel account who want an iMessage assistant that can operate real websites in a cloud browser while keeping logins and cards in a vault they host.",
      "poor_fit": "Anyone who wants a hosted assistant that works after sign-up, who is not on or ready to buy a Vercel Pro plan, or who is not comfortable letting an agent sign in and buy on their behalf.",
      "decision_note": "Self-hosted, so you carry the setup and the bills. The README says deployment needs a Vercel Pro plan, with free plans for Kernel (cloud browsers) and Neon (database) and Linq's iMessage line and Blob storage billed through Vercel; we did not verify any of those prices. The eve.dev template page for it appeared in an eve post on 5 October 2026. The vault and approval claims are the project's own and we did not audit the code. Start with approvals on and no stored cards.",
      "availability": "Open-source code on GitHub (MIT) with a Deploy with Vercel flow; the project's site is openinstinct.sh. Self-hosted in your own Vercel account; no hosted service from the project was found.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "iMessage assistant that works through real websites in a cloud browser, with screenshots and receipts sent to the thread",
          "source_url": "https://openinstinct.sh/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Free MIT software; Vercel Pro plan required to deploy",
          "region_restrictions": "Not stated"
        },
        {
          "name": "MIT-licensed source with a one-click Vercel deploy flow",
          "source_url": "https://github.com/Merit-Systems/OpenInstinct",
          "observed_at": "2026-10-23",
          "evidence_type": "repository_metadata",
          "confidence": "high",
          "plan_restrictions": "Free MIT software; Vercel Pro plan required to deploy",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Encrypted password and card vault with browser autofill that keeps secrets out of the model's context",
          "source_url": "https://openinstinct.sh/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Free MIT software; Vercel Pro plan required to deploy",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Approval in the thread before purchases, email, calendar events and destructive changes",
          "source_url": "https://openinstinct.sh/",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Free MIT software; Vercel Pro plan required to deploy",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Listed as the 'iMessage Agent as a Service' template on eve.dev",
          "source_url": "https://eve.dev/templates/openinstinct-eve-template",
          "observed_at": "2026-10-23",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Free MIT software; Vercel Pro plan required to deploy",
          "region_restrictions": "Not stated"
        }
      ],
      "permissions": [
        "Lives in your iMessage thread; works through real sites in a cloud browser including sign-in and checkout, then sends a screenshot and summary to the thread.",
        "Purchase approval: before a purchase it posts the merchant, item and total in your thread; email, calendar events and destructive changes also need approval (vendor statement).",
        "Passwords and cards are encrypted before storage; the browser fills them from the vault by handle so secrets stay out of the model.",
        "Remembers name, address, phone number and preferences for forms; forgets them on request.",
        "Works with any model via Vercel AI Gateway; usage is billed by Vercel to your own account."
      ],
      "limitations": [
        "Self-hosted: deployment needs a Vercel Pro plan, per the README",
        "Supporting services (Kernel, Neon, Linq line, Blob storage, AI Gateway) have their own free tiers, credits and usage charges, per the README; we did not verify prices",
        "Vault, browser-autofill and approval behaviour are the project's own statements; no independent audit was found",
        "Uses a cloud browser to sign in and check out on real merchant sites, which some sites restrict",
        "Not tested by AgentComparison"
      ],
      "runtime": [
        "Next.js app built on the eve agent framework, deployed to your Vercel project",
        "Kernel cloud browsers, Neon Postgres, private Vercel Blob storage, a Linq iMessage line and Vercel AI Gateway for models; optional Link wallet and Google Workspace connections"
      ],
      "privacy_security": "OpenInstinct runs inside the user's own Vercel account: the deploy flow provisions the browser, database, private Blob store and iMessage line under that account, so credentials stay inside the user's own Vercel project. Passwords and cards are encrypted before storage; the worker passes a handle to the browser, which fills the saved value from the vault - 'keeps secrets out of the model'. Saved personal details (name, address, phone, preferences) can be forgotten on request. Checked 23 October 2026.",
      "pricing": {
        "amount": 0,
        "currency": "USD",
        "billing_period": null,
        "allowances": null,
        "usage_charges": "The software is free under the MIT licence. The README says a Vercel Pro plan is required and that Linq, Blob storage and model usage are billed through Vercel; Kernel and Neon offer free plans. No total is given.",
        "verified_at": "2026-10-05",
        "source_url": "https://github.com/Merit-Systems/OpenInstinct",
        "note": "No price for the assistant itself. Running it needs a Vercel Pro plan plus metered usage for the connected services; we did not verify Vercel's prices.",
        "class": "see_vendor"
      },
      "sources": [
        "https://openinstinct.sh/",
        "https://github.com/Merit-Systems/OpenInstinct",
        "https://eve.dev/templates/openinstinct-eve-template",
        "https://x.com/eve/status/2107187302997127567"
      ],
      "changelog": [
        {
          "date": "2026-10-23",
          "change": "Field-level audit from openinstinct.sh: MIT open-source iMessage assistant in your own Vercel account; cloud-browser execution with screenshot+summary; purchase/email/calendar/destructive approvals in-thread; encrypted vault fills keeping secrets out of the model; any model via Vercel AI Gateway billed to your account; forget-on-request saved details. Coverage rows, permissions, setup steps, privacy fields filled. No hands-on testing."
        },
        {
          "date": "2026-10-05",
          "change": "Added from the project's site, GitHub repository and README, and the eve.dev template page named in an eve post. Documentation research only; not tested."
        }
      ],
      "coverage_rows": [
        [
          "Cost",
          "MIT, self-hosted, BYO usage",
          "MIT licensed and runs in your own Vercel account; Vercel bills model usage (AI Gateway) and infra to your account - no separate subscription.",
          "https://openinstinct.sh/"
        ],
        [
          "Approval model",
          "Purchase + sensitive approvals",
          "Vendor: merchant/item/total posted to thread before any purchase; email, calendar and destructive changes need approval; OTPs are asked for in-thread.",
          "https://openinstinct.sh/"
        ],
        [
          "Secret handling",
          "Encrypted vault fills",
          "Passwords/cards encrypted before storage; browser fills by handle from the vault so secrets never enter the model context (vendor statement).",
          "https://openinstinct.sh/"
        ],
        [
          "Deployment",
          "Own Vercel account",
          "Deploy flow provisions browser, database, private Blob store and iMessage line under your Vercel account; setup advertised as about five minutes plus one verification text.",
          "https://openinstinct.sh/"
        ]
      ],
      "setup_steps": [
        "Deploy to your own Vercel account (the flow provisions browser, database, Blob store and iMessage line) - review what it creates in your project first.",
        "Send the one verification iMessage, then save logins and cards in the web workspace or add them when a task asks.",
        "Run one low-stakes chore (a grocery cart build without placing it) and check the approval prompt shows merchant, item and total before approving anything."
      ],
      "price_tags": [
        "free",
        "usage"
      ],
      "price_summary": "Free MIT software; needs a Vercel Pro plan plus usage for connected services",
      "page_url": "https://agentcomparison.net/agents/openinstinct/"
    },
    {
      "id": "codex",
      "name": "OpenAI Codex",
      "vendor": "OpenAI",
      "kind": "product",
      "category": "coding agent (CLI, IDE, web and desktop app)",
      "group": "coding-agents",
      "summary": "Codex is OpenAI's coding agent. OpenAI's docs describe a CLI that inspects code, edits files and runs commands in your repository, plus an IDE extension, cloud tasks and the ChatGPT desktop app. It is included in ChatGPT plans from Free to Pro and Business, or can be used with an API key at API prices.",
      "one_liner": "OpenAI's coding agent for terminal, editor, cloud and desktop, bundled with ChatGPT plans.",
      "best_fit": "Developers who already use or are willing to buy a ChatGPT plan and want one coding agent across terminal, editor and cloud, with sandbox and approval settings they control.",
      "poor_fit": "Anyone who needs a fixed, published task allowance on the free tiers, or who needs cloud features such as GitHub code review and Slack on an API-key setup, which the pricing page says are not included.",
      "decision_note": "Codex is sold through ChatGPT plans, and the pricing page says ChatGPT Work and Codex share usage, credits and limits. OpenAI lists Free at $0, Go at $8 a month, Plus at $20, Pro from $100 (plans at $100, $200 or $500), Business at $20 per user a month billed annually ($25 monthly) and Enterprise on request. The Free and Go entries list only GPT-6 Luna in the desktop app, subject to rollout, and published usage estimates are not fixed task or message guarantees, so check what your plan allows before relying on it. The pricing page also says GPT-5.5 is scheduled to retire from Codex on 14 October 2026; this date notice does not establish the exact cutoff time.",
      "availability": "Codex CLI (Apache-2.0 on GitHub), IDE extension, Codex on the web and iOS, and the ChatGPT desktop app, per OpenAI's pricing page and docs. Which surface and model you get depends on your plan.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Terminal agent that inspects code, edits files and runs commands in your repository",
          "source_url": "https://developers.openai.com/codex/cli",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "CLI included with ChatGPT plans or an API key",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Open-source Codex CLI under the Apache-2.0 licence",
          "source_url": "https://github.com/openai/codex",
          "observed_at": "2026-10-22",
          "evidence_type": "repository_metadata",
          "confidence": "high",
          "plan_restrictions": "None for the CLI code",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Cloud integrations such as automatic code review and Slack on Plus and above",
          "source_url": "https://developers.openai.com/codex/pricing",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Not available with an API key",
          "region_restrictions": "Not stated"
        },
        {
          "name": "OS-enforced sandbox and approval policy, with network off by default",
          "source_url": "https://learn.chatgpt.com/docs/agent-approvals-security",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "All plans, admin controls on Business and Enterprise",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Codex Security: separate product that scans connected GitHub repositories",
          "source_url": "https://learn.chatgpt.com/docs/agent-approvals-security",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "OS-enforced sandbox limits what the agent can touch (typically the current workspace) plus an approval policy controlling when it must stop and ask; network access is off by default (agent approvals & security doc, re-verified 22 October 2026).",
        "The 'untrusted' approval policy is retired and can prevent startup; use read-only sandbox + on-request approvals for interactive read-only use.",
        "Current Cloud network secrets and environment access differ from Legacy Cloud setup-only secrets. Review the chosen runtime before supplying credentials."
      ],
      "limitations": [
        "Plans share usage with ChatGPT Work. Published local-message estimates depend on model and workload, not fixed task guarantees.",
        "The Free and Go entries list only GPT-6 Luna at standard speed in the desktop app, subject to rollout",
        "API-key use has no cloud features such as GitHub code review or Slack, and is billed at API prices that are separate from subscriptions",
        "Model names and availability change; OpenAI says GPT-5.5 retires from Codex on 14 October 2026",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Codex CLI (open source, Apache-2.0) that runs on your machine against a local repository",
        "Codex cloud tasks, IDE extension and the ChatGPT desktop app, depending on plan"
      ],
      "privacy_security": "Re-verified 22 October 2026 against OpenAI's current data-controls article: on personal plans, turning off 'Improve the model for everyone' in ChatGPT - or 'Do not train on my content' in the Privacy Portal - covers new Codex tasks; either one is sufficient. The separate 'Include environments' setting in Codex settings controls whether environment context can train models and is NOT changed by that opt-out. Managed Business/Enterprise/Edu content is not used for training by default; feedback and workspace policies can affect scope. No Codex-wide cloud retention period was established. Vendor documentation, not independently audited.",
      "pricing": {
        "amount": 0,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Free $0; Go $8; Plus $20; Pro from $100; Business $20 per user billed annually",
        "usage_charges": "Extra usage through ChatGPT credits on Plus, Pro and Business; API-key use is billed at API prices",
        "verified_at": "2026-10-22",
        "source_url": "https://learn.chatgpt.com/docs/pricing",
        "note": "OpenAI's Codex pricing page (now at learn.chatgpt.com/docs/pricing; read 22 October 2026) page lists Free at $0, Go at $8 a month, Plus at $20, Pro from $100 (plans at $100, $200 or $500), Business at $20 per user a month billed annually or $25 monthly with two or more users, Enterprise and Edu on request, and an API-key option billed at API prices. Usage is shared with ChatGPT Work."
      },
      "sources": [
        "https://learn.chatgpt.com/docs/pricing",
        "https://developers.openai.com/codex/cli",
        "https://learn.chatgpt.com/docs/agent-approvals-security",
        "https://github.com/openai/codex",
        "https://learn.chatgpt.com/docs/environments/cloud-environments",
        "https://learn.chatgpt.com/docs/pricing",
        "https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt"
      ],
      "changelog": [
        {
          "date": "2026-10-14",
          "change": "Expanded product-specific runtime, approval, access and privacy evidence with narrow setup steps. Documentation research, not hands-on testing."
        },
        {
          "date": "2026-10-06",
          "change": "Added from OpenAI's Codex pricing page, CLI and security docs and the public GitHub repository. Documentation research only; not tested."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: data-controls claims re-verified verbatim (Improve-the-model toggle or Privacy Portal covers Codex tasks; Include environments setting separate); sandbox/approvals doc re-read - untrusted approval policy retired, network off by default confirmed; Codex Security repo-scanning product added."
        }
      ],
      "coverage_rows": [
        [
          "Local command control",
          "Workspace and approvals",
          "Auto can read, edit and run commands within the workspace without per-command approval. Sandbox and approval policy are separate; broader permissions change the boundary. The untrusted approval policy is retired and can prevent startup.",
          "https://learn.chatgpt.com/docs/agent-approvals-security"
        ],
        [
          "Approval reviewer",
          "Human by default, configurable",
          "Requests normally route to the user. Auto-review can assess eligible escalations instead; actions inside the sandbox do not gain an extra review step. It uses additional model calls, which can add to usage.",
          "https://learn.chatgpt.com/docs/agent-approvals-security"
        ],
        [
          "Current cloud runtime",
          "Published environment",
          "Each new cloud task gets an isolated workspace from a published environment. Saving configuration, publishing prepared files and sharing access are separate steps. Cloud can continue while your computer sleeps.",
          "https://learn.chatgpt.com/docs/environments/cloud-environments"
        ],
        [
          "Current cloud secrets",
          "Direct value or proxy secret",
          "Environment variables go directly to programs. Network secrets supply placeholders, with the proxy substituting credentials only for allowed HTTPS destinations on port 443 during setup and tasks. Saving an environment-owned secret also allows its destinations.",
          "https://learn.chatgpt.com/docs/environments/cloud-environments"
        ],
        [
          "Cloud networking",
          "Review environment policy",
          "Package-manager, custom-domain and unrestricted access settings apply to current Cloud. Allowing a destination does not grant service permissions or supply credentials. Agent Security requirements and environment restrictions apply separately.",
          "https://learn.chatgpt.com/docs/environments/cloud-environments"
        ],
        [
          "Legacy cloud distinction",
          "Do not mix settings",
          "Legacy Cloud still supports Code Review and Linear/GitHub integrations. Its configured secrets are setup-only and removed before the agent phase; that restriction is not the current Cloud network-secret model.",
          "https://learn.chatgpt.com/docs/agent-approvals-security"
        ],
        [
          "Plan usage and API",
          "Estimates, not task guarantees",
          "Plus and Standard Business have model-dependent local-message estimates per five hours; Pro currently has no five-hour limit. Local/cloud share allowance and weekly limits may apply. API-key use has separate charges and no cloud GitHub code review or Slack features.",
          "https://learn.chatgpt.com/docs/pricing"
        ],
        [
          "Personal data controls",
          "Two settings to inspect",
          "Improve the model for everyone, or the Privacy Portal opt-out, covers new personal-plan Codex tasks. Include environments separately controls additional environment context and is not changed by that opt-out. Managed business workspaces default to no training; feedback can have an exception.",
          "https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt"
        ]
      ],
      "setup_steps": [
        "Choose local CLI/IDE or current Cloud first; do not copy Legacy Cloud secret rules into current Cloud.",
        "For local work, use read-only or workspace-write with on-request approval and inspect /status. Do not use the retired untrusted policy.",
        "For current Cloud, prepare one repository, review the setup report and publish a private environment before starting a new task.",
        "Give a service only the access it needs. Inspect network-secret destinations and environment restrictions; review task output before merging.",
        "On personal plans, inspect both task-training and Include environments settings, then check your usage dashboard."
      ],
      "price_tags": [
        "free",
        "listed",
        "usage"
      ],
      "price_summary": "Free $0, Go $8, Plus $20, Pro from $100 per month; Business $20/user billed annually",
      "page_url": "https://agentcomparison.net/agents/codex/"
    },
    {
      "id": "mitra",
      "name": "Mitra",
      "vendor": "Not named on the pages we read (support emails use the deusmachina.inc domain)",
      "kind": "product",
      "category": "AI phone assistant",
      "group": "personal-assistants",
      "summary": "Mitra is an AI assistant that answers and makes phone calls from your own number, and can also text, email, message on Slack, book appointments and do web tasks, according to its own site and TestingCatalog's coverage. On 5 and 6 October 2026 its founder announced Mitra Link, where your Mitra talks to other people's Mitras, and a chat takeover that lets it reply in a thread from your account.",
      "one_liner": "An AI phone assistant that answers and makes calls on your own number; plans from $30 a month.",
      "best_fit": "People who want an assistant that handles phone calls on their own number, with live listening, transfer rules and call summaries, and who accept it speaking on their behalf.",
      "poor_fit": "Anyone who needs published approval limits, a detailed security statement or independent testing before letting an assistant talk to others as them.",
      "decision_note": "Mitra lists three subscription plans at $30, $90 and $300 a month, with call hours as the main allowance, and a 7-day trial with 2 call hours. It speaks to people as you, so check transfer rules, what it may say and where your data goes before connecting accounts. Newest features (Mitra Link, chat takeover) are known only from announcements and press, not from Mitra's own help pages.",
      "availability": "Available on iPhone, iPad and Mac per the App Store listing and TestingCatalog (23 September 2026). Android and web availability not confirmed on the pages we read.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Answers and makes phone calls from your own number, with live transcript, listening in and takeover",
          "source_url": "https://www.getmitra.com/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Transfers urgent calls to you using rules you set",
          "source_url": "https://www.getmitra.com/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Post-call recording, transcript, summary and next steps",
          "source_url": "https://www.getmitra.com/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Mitra Link: your Mitra talks directly to other people's Mitras, announced 5 October 2026",
          "source_url": "https://www.testingcatalog.com/mitra-adds-chat-takeover-and-mitra-link-to-its-ai-assistant/",
          "observed_at": "2026-10-06",
          "evidence_type": "vendor_documentation",
          "confidence": "low",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Chat takeover: mention your assistant in a thread and it replies from your account, announced 6 October 2026",
          "source_url": "https://www.testingcatalog.com/mitra-adds-chat-takeover-and-mitra-link-to-its-ai-assistant/",
          "observed_at": "2026-10-06",
          "evidence_type": "vendor_documentation",
          "confidence": "low",
          "plan_restrictions": "See pricing notes",
          "region_restrictions": "Not stated"
        },
        {
          "name": "Voice cloning (may count as biometric data in some US states, per its privacy policy)",
          "source_url": "https://www.getmitra.com/privacy",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": "US state biometric laws vary"
        }
      ],
      "permissions": [
        "Mitra's site says you can set transfer rules, including whether it connects you directly or calls you first to ask, and that you can watch calls live, read the transcript and step in.",
        "Per its privacy policy, website sign-ins you approve run in a non-persistent on-device browser session restricted to that site; credentials are encrypted on-device with the key in Apple Keychain and never shown to the AI.",
        "The pricing FAQ says data-use for improving its AI systems can be turned off in settings; the privacy policy states private call content, messages, files, contacts and memories are never training data.",
        "TestingCatalog says permission settings govern access to connected apps and that Mitra Link lets you choose who can reach your Mitra. We did not find a full list of what Mitra may do without asking."
      ],
      "limitations": [
        "Pricing page lists call hours and says Mitra Link exchanges are limited, more or unlimited by plan, but gives no numbers for Link exchanges or virtual computer hours",
        "Mitra Link and chat takeover were announced 5 and 6 October 2026; we found no Mitra help page describing their controls, so details come from the founder's posts as reported by TestingCatalog",
        "Security answer on the pricing page is general (certified encryption, industry practice); we did not find a named certification, and the terms and privacy pages did not load for us",
        "Vendor legal name is not stated on the pages we read",
        "Described by the vendor; not tested by AgentComparison"
      ],
      "runtime": [
        "Calls run on Mitra's own service using your phone number; per its site it also has its own number you can add to live calls",
        "Web tasks use virtual computer hours and a login vault, per TestingCatalog's summary of the launch"
      ],
      "privacy_security": "Privacy policy (DeusMachina, Inc., updated 24 August 2026, read 22 October 2026) states private call content (audio, transcripts, summaries), messages, files, contacts and memories are NOT used to train or fine-tune AI models; no sale of personal information and no cross-context behavioural advertising. Third-party AI providers (chatbot, transcription, voice agent, voice cloning) receive data as needed to serve requests. For signed-in website actions, credentials are encrypted on your Apple device with the key kept in Apple Keychain; plaintext passwords, PINs and one-time codes are handled only on-device and not shown to the AI, and sign-in runs in a non-persistent on-device browser session restricted to the site you approved. A voice cloning feature is offered; a voice clone may count as biometric data in some US states. Account deletion via app settings or support email; data retained as needed then deleted or anonymised. Note a documentation tension: the pricing FAQ still says data is used to improve AI systems unless you opt out in settings, which is vaguer than the policy's flat no-training commitment for private content. Vendor policy, not independently audited.",
      "pricing": {
        "amount": 30,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Pro 5 call hours a month; Premier 20; Prime 80; all with unlimited parallel calls",
        "usage_charges": "Extra call hours can be bought by subscribers in Profile; prices not stated on the pricing page",
        "verified_at": "2026-10-22",
        "source_url": "https://www.getmitra.com/pricing",
        "note": "Mitra's pricing page lists Mitra Pro, Premier and Prime; the App Store listing shows in-app purchases of $30, $90 and $300 for the monthly plans and $300, $900 and $3,000 for yearly. The page says yearly saves 17% and gives new users a 7-day trial with 2 hours of calls. Virtual computer hours and Mitra Link exchanges are limited, more or unlimited by plan without figures.",
        "class": "amount"
      },
      "sources": [
        "https://www.getmitra.com/",
        "https://www.getmitra.com/pricing",
        "https://apps.apple.com/us/app/mitra-ai-phone-assistant/id6505044002",
        "https://www.testingcatalog.com/mitra-brings-ai-call-handling-to-your-existing-number/",
        "https://www.testingcatalog.com/mitra-adds-chat-takeover-and-mitra-link-to-its-ai-assistant/",
        "https://www.getmitra.com/privacy"
      ],
      "changelog": [
        {
          "date": "2026-10-06",
          "change": "Added from Mitra's site and pricing page, the App Store listing and TestingCatalog coverage (23 September and 6 October 2026). Documentation research only; no hands-on testing."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: privacy policy read for the first time (24 Aug 2026). Material correction - the policy flatly commits that call audio, transcripts, summaries, messages, files, contacts and memories are NOT used to train AI models, and documents on-device credential encryption with Apple Keychain; the record previously carried only the pricing FAQ's vaguer improve-our-AI-with-opt-out line, now flagged as a documentation tension. Voice cloning capability added."
        }
      ],
      "price_tags": [
        "listed"
      ],
      "price_summary": "$30, $90 or $300 a month (Pro, Premier, Prime), mostly by call hours; 7-day trial with 2 call hours",
      "page_url": "https://agentcomparison.net/agents/mitra/"
    },
    {
      "id": "tab",
      "name": "Tab",
      "vendor": "Terrasoft, Inc.",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Tab describes a messaging-first personal assistant that can research, create files and media, browse websites, use connected accounts and call businesses. Its homepage says payments wait for approval of the purchase, total and payee; its terms limit wallet and checkout features to eligible private-beta users.",
      "one_liner": "A private-beta assistant for tasks in Messages, calls, connected accounts and approved purchases.",
      "best_fit": "Adults exploring messaging-based personal errands, document/media work and calls, with narrow connected-account access and careful review of purchase terms.",
      "poor_fit": "Anyone needing guaranteed immediate access, a published subscription or per-task budget, independently tested outcomes, or a fixed deletion/retention timetable before connecting sensitive data.",
      "decision_note": "Consider Tab for personal admin if you can obtain beta access and accept unresolved pricing. The public terms and security statement give concrete payment and credential controls, but they are vendor claims, not an audit or proof that calls, bookings and website actions succeed.",
      "availability": "Private beta and staged invitations. The homepage requests name and mobile number, followed by one required text; joining the waitlist does not guarantee access, timing, pricing or feature availability. The terms require adulthood or the local age of legal majority. Country options in a signup form do not prove regional service eligibility.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Text, photos, screenshots, PDFs and video as task inputs; file and media creation",
          "source_url": "https://tab.bot/",
          "observed_at": "2026-10-08",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Private-beta access; features may be limited or unavailable",
          "region_restrictions": "Not established"
        },
        {
          "name": "Website browsing and requested use of connected accounts",
          "source_url": "https://tab.bot/terms",
          "observed_at": "2026-10-08",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Private-beta access; features may be limited or unavailable",
          "region_restrictions": "Not established"
        },
        {
          "name": "Outbound business calls, waiting on hold and returning a result in Messages",
          "source_url": "https://tab.bot/",
          "observed_at": "2026-10-08",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Private-beta access; features may be limited or unavailable",
          "region_restrictions": "Not established"
        },
        {
          "name": "Approved purchases and payments where wallet features are enabled",
          "source_url": "https://tab.bot/terms",
          "observed_at": "2026-10-08",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Private-beta access; features may be limited or unavailable",
          "region_restrictions": "Not established"
        },
        {
          "name": "Updates while work continues and context from earlier tasks",
          "source_url": "https://tab.bot/",
          "observed_at": "2026-10-08",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Private-beta access; features may be limited or unavailable",
          "region_restrictions": "Not established"
        },
        {
          "name": "Dated release notes describe timezone handling, browser changes and order follow-up; these are release claims, not tested outcomes",
          "source_url": "https://tab.bot/updates",
          "observed_at": "2026-10-08",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Private-beta access; features may be limited or unavailable",
          "region_restrictions": "Not established"
        },
        {
          "name": "Group chats: a separate Tab for each group, including people without their own Tab; money/password tasks move one-on-one",
          "source_url": "https://tab.bot/updates",
          "observed_at": "2026-10-08",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Private-beta availability; vendor claim, not a task test",
          "region_restrictions": "Not established"
        },
        {
          "name": "Group-to-private requests: group Tab asks the personal Tab; vendor says nothing private is shared into the group without permission",
          "source_url": "https://tab.bot/updates",
          "observed_at": "2026-10-08",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Private-beta availability; vendor claim, not a task test",
          "region_restrictions": "Not established"
        },
        {
          "name": "Reminders and recurring checks with optional active hours; vendor says it can report recent runs and check costs",
          "source_url": "https://tab.bot/updates",
          "observed_at": "2026-10-08",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Private-beta availability; vendor claim, not a task test",
          "region_restrictions": "Not established"
        },
        {
          "name": "Delivery, flight, food-order and callback follow-up with status changes and arrival checks",
          "source_url": "https://tab.bot/updates",
          "observed_at": "2026-10-08",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Private-beta availability; vendor claim, not a task test",
          "region_restrictions": "Not established"
        },
        {
          "name": "Named connected services in terms: Gmail, Calendar, Slack, GitHub, Linear and Notion, subject to selected provider scopes",
          "source_url": "https://tab.bot/terms",
          "observed_at": "2026-10-08",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Private-beta availability; vendor claim, not a task test",
          "region_restrictions": "Not established"
        },
        {
          "name": "Contacts import via a secure Apple ID credential link and a six-digit verification code, as described in September 14 release notes",
          "source_url": "https://tab.bot/updates",
          "observed_at": "2026-10-08",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Private-beta availability; vendor claim, not a task test",
          "region_restrictions": "Not established"
        },
        {
          "name": "Spreadsheets, PDFs, 3D files, ZIPs and published pages under sites.tab.bot, per September 11 release notes",
          "source_url": "https://tab.bot/updates",
          "observed_at": "2026-10-08",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Private-beta availability; vendor claim, not a task test",
          "region_restrictions": "Not established"
        },
        {
          "name": "Outbound calls with up to an hour on hold, per September 9 release notes; duration is a vendor statement, not a measured guarantee",
          "source_url": "https://tab.bot/updates",
          "observed_at": "2026-10-08",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Private-beta availability; vendor claim, not a task test",
          "region_restrictions": "Not established"
        }
      ],
      "permissions": [
        "A request authorizes steps reasonably necessary within the supplied scope, according to the terms. Ambiguous, sensitive, irreversible or materially changed requests may be paused for clarification; this is not a promise of a new approval for every action.",
        "Connected services use the access shown by the provider. The security page says write actions are denied by default unless the exact action is enabled; review the actual connection scope and settings.",
        "The homepage says spending waits until the user sees the purchase, total and payee and approves. Terms additionally bind merchant, cart, recipient, currency, payment destination and a short deadline; changed terms or uncertain outcomes are designed to stop, not silently retry.",
        "Payment and wallet features are limited beta. Bank or cardholder verification may require the user to take over. A successful demo does not establish that a particular merchant or payment works.",
        "Outbound calls and communications must be lawful and have any required permission. Tab may identify itself; the terms prohibit spam and emergency-service use.",
        "Card enrollment uses a single-purpose private link that expires after 10 minutes, per the security statement. Full card details must not be sent in Messages; this is a vendor-described control, not a test.",
        "Group work does not imply consent to expose private context: release notes say money/password jobs move into the personal thread and group-to-personal requests require permission before private data is shared.",
        "Apple contacts import is described as requiring saved Apple ID credentials through a secure link and a verification code. This is broader than uploading a contacts file; verify the actual scope, revocation and account-security implications first.",
        "Terms cap total liability at the greater of $100 or service fees paid in the preceding 12 months, subject to law. The reviewed terms contain no arbitration clause; that is a description of this version, not legal advice."
      ],
      "limitations": [
        "No official subscription price, per-task fee, usage allowance or numeric service limit was found on these pages. Unknown price does not mean free.",
        "Joining the waitlist does not guarantee beta access; calls, connections, media generation and payments may be incomplete, limited or unavailable.",
        "No hands-on test of task completion, call quality, response time, spending controls or reliability was run. No certification or independent security audit is claimed.",
        "The privacy policy gives purpose-based retention with legal, security, backup and fraud-prevention exceptions, not a fixed deletion window. Disconnecting account access is not an explicit promise to erase retained task data.",
        "No explicit model-training exclusion was established in the official privacy policy, terms or security page reviewed. A founder quote in press reporting is not a substitute for the applicable policy.",
        "The security page was reviewed 21 August and terms/privacy took effect 2 September, earlier than the October launch reporting. Confirm current beta behavior and terms before relying on them.",
        "Terms name Gmail, Calendar, Slack, GitHub, Linear and Notion. The competitor article also lists Drive, Docs, Sheets, Slides, Contacts and Tasks through one Google sign-in, but that exact bundle was not established by the official pages reviewed. Confirm the actual consent screen rather than assuming it.",
        "Microsoft 365/Outlook support was not established in the official pages reviewed. Absence from these pages is not proof that it is unsupported.",
        "Release notes dated September 10 say everyone wanting access gets in; the current homepage still says private beta/waitlist and the terms reserve staged access. Treat the current entry form and terms as controlling, not the older availability claim.",
        "The privacy notice confirms verified invite attribution and that neither participant receives the other's phone number. The competitor's claim that referrals move users up the waitlist was not established by the official pages reviewed."
      ],
      "runtime": [
        "Vendor-managed Messages experience with a browser and computer for requested tasks",
        "Messages and WhatsApp; official release notes document WhatsApp support and media on 10-11 September. iMessage is additionally reported by TechCrunch.",
        "Connected-service and wallet features depend on beta enablement; no account was created or tested"
      ],
      "privacy_security": "Vendor policy: messages, files, call results, browser activity and requested transaction data are processed to provide the service. Calls may include recordings or transcripts when configured and applicable consent requirements are addressed. Tab says it does not sell personal information. Its security statement describes encrypted saved credentials, private card enrollment and scoped account access; these are vendor statements, not an independent audit. Retention is as long as reasonably needed, with legal, security, backup and fraud-prevention exceptions and no fixed timetable. Users can ask to disconnect accounts, revoke credentials or request access, correction and deletion; identity checks may apply. The reviewed official policies do not establish a blanket no-model-training promise. Third-party services apply their own terms and privacy practices.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-23",
        "source_url": "https://tab.bot/terms",
        "note": "Re-checked 23 October 2026: still no public subscription amount, usage allowance or per-task rate on tab.bot homepage, terms, privacy, security or updates pages.",
        "class": "waitlist"
      },
      "sources": [
        "https://tab.bot/",
        "https://tab.bot/terms",
        "https://tab.bot/privacy",
        "https://tab.bot/security",
        "https://tab.bot/updates",
        "https://techcrunch.com/2026/10/07/another-personal-ai-assistant-has-launched-meet-tab-which-emerged-from-stealth-with-a-300m-valuation/"
      ],
      "changelog": [
        {
          "date": "2026-10-08",
          "change": "Added after identifying the exact linked X topic as Tab. Read five official pages and separated private-beta/payment controls from unverified pricing, training and retention claims. No signup or hands-on test."
        },
        {
          "date": "2026-10-08",
          "change": "Expanded feature-by-feature from official terms, security and dated release notes: groups/private threads, recurring checks, delivery follow-up, named integrations, Apple contacts, files/web publishing, hold time, signup steps, card-link expiry and a Tab/Carly decision table. Flagged full Google bundle, Microsoft support and referral-ranking gaps rather than guessing. No new title/meta or tested outcomes."
        },
        {
          "date": "2026-10-23",
          "note": "Fresh source-to-field re-check (phase 2): payment approval flow re-verified ('Nothing is paid until you approve it; Tab pays, sends the receipt back'); security page re-read - AES-256-GCM for stored passwords, opaque card references only, keyed-hash session tokens, masked secret inputs on screenshots; privacy - no sale, limited session metadata retention, delete/de-identify when no longer needed. Record statements still accurate. Counts toward the detailed ledger again."
        }
      ],
      "coverage_rows": [
        [
          "Group chat and private threads",
          "Documented",
          "Sept 10-11 release notes describe separate group agents, private money/password handling and permission before group disclosure.",
          "https://tab.bot/updates"
        ],
        [
          "Recurring checks and reminders",
          "Documented",
          "Sept 5 notes cover time windows, recent-run status and recurring-check cost reporting. No rate or delivery reliability was tested.",
          "https://tab.bot/updates"
        ],
        [
          "Named integrations",
          "Partly established",
          "Terms name Gmail, Calendar, Slack, GitHub, Linear and Notion. Exact full Google bundle and Microsoft availability remain unverified.",
          "https://tab.bot/terms"
        ],
        [
          "Contacts import",
          "Documented, sensitive",
          "Sept 14 notes describe Apple ID credentials via a secure link plus a code. Check account scope first.",
          "https://tab.bot/updates"
        ],
        [
          "Calls and hold time",
          "Documented",
          "Calls/hold are on the homepage; Sept 9 notes claim up to one hour. No call was tested.",
          "https://tab.bot/updates"
        ],
        [
          "Files and web publishing",
          "Documented",
          "Sept 11 notes add 3D files, ZIPs and pages under sites.tab.bot to homepage file/media claims.",
          "https://tab.bot/updates"
        ],
        [
          "Signup and referrals",
          "Partly established",
          "Homepage requires name, mobile and one text. Privacy confirms verified invite attribution and number privacy; queue ranking is unverified.",
          "https://tab.bot/privacy"
        ],
        [
          "Purchase approval and card expiry",
          "Documented, limited beta",
          "Security gives a 10-minute card link and transaction-bound controls; terms do not promise universal wallet access.",
          "https://tab.bot/security"
        ],
        [
          "Model training",
          "Policy gap",
          "Written policy does not establish a blanket training exclusion. No founder quote is promoted to policy.",
          "https://tab.bot/privacy"
        ],
        [
          "Retention and calls",
          "Documented, no fixed window",
          "Purpose-based retention, consent/configuration-dependent recording and request-based deletion; no fixed deletion SLA.",
          "https://tab.bot/privacy"
        ]
      ],
      "setup_steps": [
        "Open tab.bot and read the current private-beta terms. Eligibility is at least 18 or the age of majority where you live; country choices do not guarantee service availability.",
        "Enter your own name and mobile number with the correct country code. The homepage asks for automated waitlist/product-text consent and describes STOP/HELP controls. No submission was made for this review.",
        "Complete the required text step shown by Tab. Its homepage says your number plus one required text finishes signup; official updates document WhatsApp as another channel. Follow the actual on-screen route, not a guessed phone number.",
        "Wait for actual access before granting connections. Review the provider's account/workspace scopes, connect only what your first task needs, and ask for a draft or low-stakes file before allowing sends or spending.",
        "Keep passwords and card details out of the message thread. Use the vendor's private credential/card flow; the security page says the card-enrollment link expires in 10 minutes. Review item, total, payee and merchant terms before any purchase."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Private beta; public service price and usage allowance not established",
      "extra_faq": [
        {
          "question": "Is Tab at tab.bot the same as Cursor Tab or the former Tab pendant?",
          "answer": "No. This page is about the Terrasoft messaging assistant at tab.bot. Cursor's official docs call Cursor Tab an AI code autocomplete feature. The Verge reported in July 2024 that Avi Schiffmann's pendant was initially named Tab before becoming Friend. Those are different products and their plans are not prices for this texting agent.",
          "sources": [
            "https://cursor.com/docs/tab/overview",
            "https://www.theverge.com/2024/7/30/24207029/friend-ai-companion-gadget"
          ]
        }
      ],
      "meta_title": "Tab assistant (Oct 2026): private beta, no public price",
      "page_url": "https://agentcomparison.net/agents/tab/"
    },
    {
      "id": "carly",
      "name": "Carly",
      "vendor": "SWH LABS LLC",
      "kind": "product",
      "category": "business assistant",
      "group": "work-business",
      "summary": "Carly, formerly CalBot, documents an email-first assistant and custom agents for Gmail and Outlook, calendar scheduling, CRM work, research and event-triggered workflows. Users can delegate by email, SMS, Slack or dashboard chat, or use its MCP connector. The paid assistant and team agents are distinct from its limited free workflow and MCP tools; business calling and the AI receptionist are separately billed add-ons.",
      "one_liner": "Email, calendar and CRM agents with event-triggered workflows, Google and Microsoft connections, and scoped team access.",
      "best_fit": "Individuals or teams delegating email, multi-calendar scheduling, CRM follow-up and recurring workflows across Google, Microsoft and connected business apps.",
      "poor_fit": "Anyone needing independently tested reliability, a general-purpose personal shopping wallet, phone service included in the base subscription, or unrestricted free autonomous email sending.",
      "decision_note": "Consider Carly when the job is inbox, calendar, CRM or a repeatable business workflow rather than a general shopping errand. Its published prices and Google/Microsoft support help narrow the fit. Read the selected agent tools and sending mode before connecting accounts: documented scopes and MCP approval advice do not establish a universal approval before every email or automation.",
      "availability": "Public signup and a Free plan are advertised, alongside Personal, Organization and Enterprise. No account was created. MCP documentation says no region restrictions; that statement does not establish that business phone numbers, calling or every integration work in every country.",
      "last_checked": "2026-10-23",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Gmail and Outlook email delegation, drafts and sending through selected account or agent address",
          "source_url": "https://www.usecarly.com/blog/how-to/how-to-get-started-carly-agent/",
          "observed_at": "2026-10-09",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Selected plan and connected scopes; sending and phone features have paid restrictions",
          "region_restrictions": "MCP says no region restrictions; phone eligibility not established"
        },
        {
          "name": "Multi-calendar scheduling, booking pages, availability polls and meeting preferences",
          "source_url": "https://www.usecarly.com/pricing/",
          "observed_at": "2026-10-09",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Selected plan and connected scopes; sending and phone features have paid restrictions",
          "region_restrictions": "MCP says no region restrictions; phone eligibility not established"
        },
        {
          "name": "Custom email agents with individual instructions, tool scopes and knowledge files",
          "source_url": "https://www.usecarly.com/blog/how-to/how-to-get-started-carly-agent/",
          "observed_at": "2026-10-09",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Selected plan and connected scopes; sending and phone features have paid restrictions",
          "region_restrictions": "MCP says no region restrictions; phone eligibility not established"
        },
        {
          "name": "Event-triggered workflows for emails, calendars, webhooks, schedules and connected apps",
          "source_url": "https://www.usecarly.com/workflows/",
          "observed_at": "2026-10-09",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Selected plan and connected scopes; sending and phone features have paid restrictions",
          "region_restrictions": "MCP says no region restrictions; phone eligibility not established"
        },
        {
          "name": "CRM updates, lead follow-up, research, to-dos and briefings",
          "source_url": "https://www.usecarly.com/pricing/",
          "observed_at": "2026-10-09",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Selected plan and connected scopes; sending and phone features have paid restrictions",
          "region_restrictions": "MCP says no region restrictions; phone eligibility not established"
        },
        {
          "name": "MCP read/write tools for compatible clients, with sensitive-write classification and API-key scopes",
          "source_url": "https://www.usecarly.com/mcp/reference/",
          "observed_at": "2026-10-09",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Selected plan and connected scopes; sending and phone features have paid restrictions",
          "region_restrictions": "MCP says no region restrictions; phone eligibility not established"
        },
        {
          "name": "SMS, Slack and dashboard chat channels",
          "source_url": "https://www.usecarly.com/pricing/",
          "observed_at": "2026-10-09",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Selected plan and connected scopes; sending and phone features have paid restrictions",
          "region_restrictions": "MCP says no region restrictions; phone eligibility not established"
        },
        {
          "name": "Organization shared agents, with separately billed business calling and AI receptionist add-ons",
          "source_url": "https://www.usecarly.com/pricing/",
          "observed_at": "2026-10-09",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Selected plan and connected scopes; sending and phone features have paid restrictions",
          "region_restrictions": "MCP says no region restrictions; phone eligibility not established"
        }
      ],
      "permissions": [
        "Custom email agents have their own instructions, email address and individually enabled tool access. The setup guide says they can only use the tools switched on; connections can be revoked.",
        "Sending mode can use the agent address or the user's connected Gmail/Outlook account. Review recipient, sender identity and automation scope before enabling live sends. No blanket per-message approval guarantee was established.",
        "MCP reference separates read, write and sensitive-write tools. Sending, deleting, running code and starting ongoing workflows are sensitive writes; it advises client approval for each call. OAuth acts for the authorizing Carly account; scoped API keys are an alternative.",
        "Calendar, mailbox, contacts and Drive access depend on the connections selected. The privacy notice distinguishes CC-thread access for core scheduling from optional broader mailbox and Drive access for CRM/agents.",
        "Business WhatsApp features operate at the connected business's direction. The business remains responsible for message content, recipients, consent and configured workflows; Meta limits and charges still apply.",
        "Paid subscription renewal is automatic unless cancelled. Credit top-ups are prepaid with no automatic charge when exhausted, according to the pricing FAQ; these are different billing mechanisms."
      ],
      "limitations": [
        "No hands-on test of email quality, scheduling accuracy, workflow reliability, response speed or approval enforcement was run. Testimonials and vendor response-time claims are not test results.",
        "Free plan lists 10 Zapier-style workflows. Non-AI steps are described as free to run, but that is not unlimited free AI assistance or paid-email sending. MCP reference says sending email requires a paid plan and some workflow features do too.",
        "Personal includes 200 credits/month in the live plan card. Numeric Organization pooled allowance, exact general task-to-credit conversion, phone/receptionist prices and Enterprise terms were not established.",
        "Pricing selector offers larger monthly credit allowances for additional fees. The FAQ describes a 7,500-credit add-on while the selector shows 7,700 total including 200 base; do not confuse extra credits with total allowance.",
        "Business SMS/calling and AI receptionist appear in the Organization feature list, but the FAQ says they are separately billed add-ons that require setup. No phone price or country eligibility is confirmed here.",
        "The policy claims encryption, role-based controls, security assessments and penetration tests; no independent audit report or certification was verified.",
        "Disconnecting an integration stops access but does not by itself erase stored records. Organization records and copies retained by recipients or other services have separate deletion rules."
      ],
      "runtime": [
        "Vendor-hosted dashboard plus email delegation by send, CC or forward",
        "SMS and Slack messaging; connected Gmail or Outlook sending can use the user's own address",
        "MCP access from Claude, ChatGPT and compatible clients; separate booking API and CLI documented",
        "Workflow triggers include email, calendar events, schedules, webhooks and connected-app events; organization agents can be shared"
      ],
      "privacy_security": "Carly's 7 October 2026 privacy notice explicitly says customer data, including Google, Microsoft and WhatsApp data, is not used to train AI or machine-learning models. It says contractual terms prevent AI providers from retaining user data or using it for training. This is a vendor policy, not an independently checked provider contract. It distinguishes calendar/contact access and optional broader email/Drive permissions; business WhatsApp may include messages, attachments and permitted history imports. Account-associated personal data is deleted within 90 days after account deletion, with legally required retention exceptions and separate rules for continuing organization accounts and copies held independently. Deletion/correction can be requested from the Account page or support; identity/authority checks may apply. Encryption, RBAC and security assessments are stated, not independently audited. Business customers can request the DPA and subprocessor list.",
      "pricing": {
        "amount": 35,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Personal: 200 credits/month; Free: 10 workflows plus limited MCP tools",
        "usage_charges": "Optional additional credit allowances and separately billed business phone/receptionist add-ons",
        "verified_at": "2026-10-23",
        "source_url": "https://www.usecarly.com/pricing/",
        "note": "Free $0/month advertises MCP access, 10 Zapier-style workflows, multi-calendar view and Task Manager. Personal is $35/month with 200 credits/month; Organization is $35/seat/month with a pooled allowance whose amount was not established; Enterprise is custom. Live Personal credit selector: 400 total +$30/month, 700 +$75, 2,200 +$260, 5,200 +$575, 7,700 +$825, 10,200 +$1,050, 15,200 +$1,450, each on top of the base plan. FAQ says deterministic non-LLM steps and MCP tool calls consume no Carly credits; an LLM workflow call costs 0.025 credits. Free MCP sending still requires a paid plan. General assistant task costs have no exact formula; recurring scheduled tasks may consume more. Credit exhaustion does not trigger an automatic top-up, but paid subscriptions auto-renew. Business SMS/calling and AI receptionist are separately billed Organization add-ons; phone and Enterprise prices remain unknown. Terms say cancellation takes effect immediately without a refund. Confirm taxes, selected allowance and third-party/Meta charges before paying."
      },
      "sources": [
        "https://www.usecarly.com/",
        "https://www.usecarly.com/pricing/",
        "https://calbotservice.com/privacy",
        "https://calbotservice.com/terms",
        "https://www.usecarly.com/workflows/",
        "https://www.usecarly.com/blog/how-to/how-to-get-started-carly-agent/",
        "https://www.usecarly.com/mcp/reference/",
        "https://www.usecarly.com/sms/",
        "https://www.usecarly.com/use-cases/"
      ],
      "changelog": [
        {
          "date": "2026-10-09",
          "change": "Re-read official features, permissions, pricing and privacy sources; added product-specific evidence table and narrow-start steps. Vendor documentation, no hands-on tests."
        },
        {
          "date": "2026-10-08",
          "change": "Added from nine official pages, live pricing cards and the current October 7 privacy/terms. Separated free tools from paid sending, credit top-ups from auto-renewal, and phone add-ons from seat price. Documentation researched, not tested."
        },
        {
          "date": "2026-10-23",
          "note": "Fresh source-to-field re-check (phase 2): all tiers re-verified - Free $0 (MCP access, 10 workflows, multi-calendar, Task Manager), Personal $35/mo with 200 credits included, Organization $35/seat/mo, credit add-ons 400 +$30 to 15,200 +$1,450. Added coverage row: agent-sent emails and no-LLM deterministic workflows cost no credits. Counts toward the detailed ledger again."
        }
      ],
      "coverage_rows": [
        [
          "Email/calendar agents",
          "Documented",
          "Google and Microsoft connections; agent address or selected own inbox sending; scheduling and multi-calendar work.",
          "https://www.usecarly.com/"
        ],
        [
          "Custom agents and scopes",
          "Documented",
          "Per-agent instructions, integrations, permissions and knowledge. Check enabled tools and sender mode before live sends.",
          "https://www.usecarly.com/blog/how-to/how-to-get-started-carly-agent/"
        ],
        [
          "Scheduled/event workflows",
          "Documented",
          "Email/calendar/app events, schedules and webhooks; deterministic non-LLM steps free, model workflow calls 0.025 credits. General assistant cost has no exact formula.",
          "https://www.usecarly.com/pricing/"
        ],
        [
          "MCP read/write",
          "Plan restriction",
          "Read/organize/draft tools have free access; sending needs paid plan. Sensitive-write classifications advise client approval, not a universal runtime guarantee.",
          "https://www.usecarly.com/mcp/reference/"
        ],
        [
          "Teams and business phone",
          "Separate price",
          "Organization shared agents and pooled credits; SMS/calling/receptionist separately billed add-ons, not included simply because listed on seat card.",
          "https://www.usecarly.com/pricing/"
        ],
        [
          "Free/paid allowances",
          "Published",
          "Free 10 workflows and limited tools; Personal $35/mo with 200 credits, Organization $35/seat/mo, Enterprise custom. Optional extra allowances and external costs apply.",
          "https://www.usecarly.com/pricing/"
        ],
        [
          "Training/security",
          "Vendor policy",
          "No model training on user data, including Google/Microsoft/WhatsApp; provider contractual no-retention/training stated. No provider-contract audit performed.",
          "https://calbotservice.com/privacy"
        ],
        [
          "Deletion and revocation",
          "Exceptions",
          "Account-associated deletion within 90 days after account deletion, legal exceptions; organization/recipient/Meta records differ. Disconnect is not erasure.",
          "https://calbotservice.com/privacy"
        ],
        [
          "Subscription vs credits",
          "Different billing",
          "Paid subscription auto-renews; cancellation immediate/no refund per terms. Credit exhaustion does not auto-top-up, per pricing FAQ.",
          "https://calbotservice.com/terms"
        ],
        [
          "No-credit actions",
          "Some actions are free",
          "Carly's pricing FAQ: agent-sent emails and deterministic workflows without LLM calls cost no credits, so continuous triggers (label and log every GitHub email, a 7am HN digest) can run free.",
          "https://www.usecarly.com/pricing/"
        ]
      ],
      "setup_steps": [
        "Review the selected Free, Personal or Organization plan and any optional credit allowance or business-phone quote.",
        "Follow the vendor setup guide; create an agent with only the integrations and tools needed for one draft or calendar task.",
        "Review sender identity and tool permissions before allowing live sending or scheduled workflows. MCP sensitive-write approval guidance depends on your client.",
        "Check recurring-task credit burn and distinguish free deterministic steps from AI calls. Phone add-ons and third-party/Meta costs are separate.",
        "Revoke unused connections and request deletion separately. If cancelling a paid subscription, read its immediate cancellation/no-refund terms first."
      ],
      "price_tags": [
        "free",
        "listed",
        "usage"
      ],
      "price_summary": "Free limited tools; Personal $35/month; Organization $35/seat/month; credits and phone add-ons",
      "page_url": "https://agentcomparison.net/agents/carly/"
    },
    {
      "id": "boba",
      "name": "Boba",
      "vendor": "humans& ai, inc.",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Boba documents a 'social intelligence' assistant that works one-on-one and inside group chats in Apple Messages and now on the web, covering goals, planning, answers, drafting and creating, with no download or login.",
      "one_liner": "An assistant that joins your iMessage one-on-one and group chats, now also on the web, free per its homepage.",
      "best_fit": "Group planning and quick answers inside an existing iMessage thread.",
      "poor_fit": "Anyone unwilling to have message content used for model training under the current privacy notice.",
      "decision_note": "Boba's homepage claims the service is free, but no pricing page or paid-plan terms were established, so treat long-term cost as unconfirmed. Its privacy notice (21 October 2026) permits training and fine-tuning on user Input and Output; Google User Data is excluded from generalised training. Review that trade-off before adding it to busy group chats.",
      "availability": "Vendor describes use inside the Apple Messages app, one-on-one and in group chats, and a new Boba Web version (free, no app, no sign-up). Signup and delivery reliability were not tested.",
      "last_checked": "2026-10-22",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "One-on-one and group-chat assistance in Apple Messages",
          "source_url": "https://boba.app",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Boba Web assistant (free, no app, no sign-up)",
          "source_url": "https://boba.app",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Group planning, search and answers",
          "source_url": "https://boba.app",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Drafting and creating (for example, a song from shared memories)",
          "source_url": "https://boba.app",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "low",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Boba participates in one-on-one threads and any group chat you add it to; the homepage says you can remove it at any time.",
        "The privacy notice references Google account connections via Google APIs; revoke access in Boba's settings or your Google account permissions."
      ],
      "limitations": [
        "The free claim appears on the homepage; no pricing page or future paid-plan terms were found.",
        "The privacy notice (5 October 2026) says Input and Output may be used to train or fine-tune the vendor's models.",
        "Reliability, response speed and behaviour in large group chats were not tested.",
        "A general fixed retention period for message content was not established in the sections reviewed."
      ],
      "runtime": [
        "Vendor-managed assistant inside Apple Messages"
      ],
      "privacy_security": "Privacy notice last updated 21 October 2026. It permits use of Input and Output to improve products and technology, including training or fine-tuning the models behind the service. Google User Data gets stricter treatment: not used for advertising or to train models beyond the user's personalised model, encrypted in transit and at rest, deletable on request to hello@boba.app within 60 days. The notice says the vendor takes commercially reasonable security steps but does not promise security.",
      "pricing": {
        "amount": 0,
        "currency": "USD",
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-22",
        "source_url": "https://boba.app",
        "note": "Free per homepage (read 22 October 2026): 'Free - No app - No sign-up'. No pricing page or paid plan was found, so future cost is unconfirmed.",
        "class": "free"
      },
      "sources": [
        "https://boba.app",
        "https://boba.app/privacy"
      ],
      "changelog": [
        {
          "date": "2026-10-20",
          "change": "Initial documentation research from homepage and privacy notice. No hands-on testing."
        }
      ],
      "coverage_rows": [
        [
          "Messaging channels",
          "Documented",
          "Apple Messages (iMessage), one-on-one and group chats, plus Boba Web on boba.app; the vendor says no download and no login.",
          "https://boba.app"
        ],
        [
          "Group behaviour",
          "Documented",
          "Vendor says it hangs back during normal chatter and jumps in when needed, and that you can remove it at any time.",
          "https://boba.app"
        ],
        [
          "Planning, answers, drafting",
          "Documented",
          "Examples include finding cabins, settling arguments and turning shared memories into a song. Output quality not tested.",
          "https://boba.app"
        ],
        [
          "Cost",
          "Homepage claim",
          "'$0, no card' on the homepage; no pricing page or paid-plan terms found.",
          "https://boba.app"
        ],
        [
          "Training and retention",
          "Policy reviewed",
          "Input and Output may be used for model training and fine-tuning. Google User Data is excluded from generalised training and deletable within 60 days of request.",
          "https://boba.app/privacy"
        ]
      ],
      "setup_steps": [
        "Text Boba from the Apple Messages app, then add it to a group chat only if you want it there.",
        "Read the privacy notice first: it permits training on your messages.",
        "Keep sensitive topics out of threads it can read until you are comfortable with its data use.",
        "Remove it from any group where members have not agreed to its presence."
      ],
      "price_tags": [
        "free"
      ],
      "price_summary": "Free per homepage ('Free - No app - No sign-up'); no pricing page established",
      "page_url": "https://agentcomparison.net/agents/boba/"
    },
    {
      "id": "boski",
      "name": "Boski",
      "vendor": "Boski, Inc.",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Boski documents a personal AI assistant reachable on Apple Messages, WhatsApp, X (XChat), e-mail and its web and mobile apps, with built-in context-sharing apps (notes, todos, journal, news, media, contacts, calorie tracking) and proactive outreach. The homepage pitches call-making for bills, bookings and job applications.",
      "one_liner": "A cross-channel personal assistant on Apple Messages, WhatsApp, X, e-mail and its apps, with built-in life-admin apps.",
      "best_fit": "A cross-channel personal assistant with proactive reminders and built-in life-admin apps.",
      "poor_fit": "Users who need published pricing before signing up, or verified hands-on reliability.",
      "decision_note": "Boski's privacy policy is unusually specific for the category: it says User Content is not used to train its own or third-party foundation models, and it gives concrete retention windows (provider cleanup within 30 days of account deletion; connected-service data deleted within 30 days of disconnection). No public pricing was found, so confirm cost before relying on it.",
      "availability": "Vendor documents the assistant across Apple Messages, WhatsApp, X (XChat) and e-mail, plus web and mobile apps, staying the same assistant across them. Signup and reliability were not tested.",
      "last_checked": "2026-10-24",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Assistant on Apple Messages, WhatsApp, X, e-mail, web and app",
          "source_url": "https://boski.com",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Built-in apps with shared context (notes, todos, journal, news, media, contacts, calories)",
          "source_url": "https://boski.com",
          "observed_at": "2026-10-20",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Proactive reminders and outreach",
          "source_url": "https://boski.com",
          "observed_at": "2026-10-20",
          "evidence_type": "vendor_documentation",
          "confidence": "low",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Memory across conversations",
          "source_url": "https://boski.com",
          "observed_at": "2026-10-20",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "You can disconnect a connected service or revoke authorisation at any time through Boski or the connected service; disconnection stops new collection but does not delete the account.",
        "Outward-facing actions (sending messages, updating external records, sharing files) happen at the user's direction per the privacy policy; a universal per-action approval gate was not established."
      ],
      "limitations": [
        "Terms (effective 20 August 2026) prohibit using the service or non-public outputs to develop, train or benchmark a competing product for publication, except where mandatory law permits.",
        "No public pricing page was found on the pages reviewed (re-checked 24 October 2026).",
        "Proactive examples (booking a haircut, chasing a debt) are vendor marketing, not verified behaviour.",
        "Reliability, response speed and the quality of the built-in apps were not tested.",
        "The vendor may enrich user profiles from public professional or social sources; users can ask for correction or deletion."
      ],
      "runtime": [
        "Vendor-managed assistant across messaging channels, web and app"
      ],
      "privacy_security": "Privacy policy effective 4 September 2026 (Boski, Inc., Delaware). It states User Content is not used for advertising and not used to train the vendor's own or third-party foundation models, and AI providers may not use it to improve general-purpose models; narrowly scoped internal systems (such as a model-selection router) may use it. Retention: account deletion starts immediately with provider cleanup ordinarily within 30 days; connected-service data deleted within 30 days of disconnection; product analytics kept at most one year. The vendor may search public sources to enrich a user profile; correction or deletion can be requested.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-24",
        "source_url": "https://boski.com",
        "note": "No pricing page found on the site as read on 24 October 2026; cost is not established.",
        "class": "see_vendor"
      },
      "sources": [
        "https://boski.com",
        "https://boski.com/privacy",
        "https://boski.com/terms"
      ],
      "changelog": [
        {
          "date": "2026-10-20",
          "change": "Initial documentation research from homepage and privacy policy. No hands-on testing."
        },
        {
          "date": "2026-10-24",
          "change": "Re-verified privacy policy clauses (no foundation-model training, 30-day retention windows, public-source enrichment). Added terms page (effective 20 Aug 2026) which bars benchmarking a competing product. Channels updated to current sources: e-mail added, Safari no longer shown. Pricing still not published."
        }
      ],
      "coverage_rows": [
        [
          "Messaging channels",
          "Documented",
          "Apple Messages, WhatsApp, X (XChat) and e-mail, plus web and mobile apps, per the privacy policy; same assistant across channels.",
          "https://boski.com"
        ],
        [
          "Built-in apps",
          "Documented",
          "Notes, todos, journal, news, media, contacts and calorie tracking share context; users can build missing apps.",
          "https://boski.com"
        ],
        [
          "Proactive behaviour",
          "Vendor examples",
          "Vendor shows proactive birthday, booking and debt-reminder examples; real behaviour not tested.",
          "https://boski.com"
        ],
        [
          "Cost",
          "Not established",
          "No public pricing found on the pages reviewed.",
          "https://boski.com"
        ],
        [
          "Training and retention",
          "Policy reviewed",
          "No foundation-model training on User Content; 30-day cleanup windows after deletion or disconnection; analytics kept at most one year.",
          "https://boski.com/privacy"
        ]
      ],
      "setup_steps": [
        "Start on one channel (for example iMessage) and confirm the assistant responds before adding more.",
        "Read the privacy policy's enrichment clause: Boski may look you up in public sources.",
        "Connect one low-risk service first and check what it can access.",
        "Ask about current pricing before depending on it, since none is published."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Not published on the pages we read",
      "page_url": "https://agentcomparison.net/agents/boski/"
    },
    {
      "id": "brea",
      "name": "Brea",
      "vendor": "Brea AI LLC",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Brea documents hosted personal AI agents via its website, web dashboard and iOS app: scheduled workflows, app integrations, price and episode watching, browser automation that fills forms and books on websites.",
      "one_liner": "Hosted personal agents with integrations, schedules and browser automation, free per its homepage.",
      "best_fit": "Scheduled personal automations (inbox summaries, price watches) and browser tasks across many integrations.",
      "poor_fit": "Saving site logins for an agent before reviewing its approval behaviour.",
      "decision_note": "Brea's homepage now says 'Free, with every feature included' and its pricing page redirects there; the earlier $9 Starter tier is no longer published. Browser automation with saved logins remains the capability to weigh carefully: scope what it can reach before connecting accounts.",
      "availability": "Website, web dashboard and iOS app per the privacy policy (updated 15 October 2026). In-app purchases are processed by Apple. Reliability was not tested.",
      "last_checked": "2026-10-24",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Scheduled workflows (for example a morning inbox summary)",
          "source_url": "https://brea.ai",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "App integrations",
          "source_url": "https://brea.ai",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Browser automation (forms, price comparison, booking)",
          "source_url": "https://brea.ai",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Price and episode watching with text alerts",
          "source_url": "https://brea.ai",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "low",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Integrations connect per app; Google integrations can be disconnected from the dashboard or iOS app, which deletes stored credentials, or revoked at myaccount.google.com/permissions.",
        "Saved site logins are stored encrypted, used only to sign in to that site, and can be replaced or deleted in Settings. A universal per-action approval gate was not established."
      ],
      "limitations": [
        "Browser automation quality and safety on real sites were not tested.",
        "Deleting your account does not cancel an App Store subscription; that must be cancelled in iOS settings.",
        "Notifications can show the first 180 characters of a reply on the lock screen, depending on iOS settings.",
        "brea.ai runs Meta and OpenAI advertising pixels; on signup or subscription they receive hashed email, name and billing location (privacy policy, 23 October 2026)."
      ],
      "runtime": [
        "Vendor-hosted agents on website, web dashboard and iOS app"
      ],
      "privacy_security": "Privacy policy last updated 23 October 2026 (Brea AI LLC). Sign-in is via Apple or Google. Conversations and files are stored so the agent can work and are deleted when the account is deleted. Google user data is used only for requested features, is not used to develop, improve or train generalised AI/ML models, and model providers may not use it for training. Saved site logins are encrypted before storage and used only to sign in to that site. Named advertising trackers run on brea.ai: the Meta Pixel plus Conversions API and an OpenAI Ads measurement pixel plus Conversions API; on account and subscription events Meta receives hashed email, name and billing city, state and postal code, and OpenAI receives hashed email, account identifiers and billing city, country and postal code. Service providers include messaging delivery for SMS, iMessage, WhatsApp and Telegram. Account deletion removes agents and website-bought subscriptions but not App Store subscriptions; stored Google data can be deleted by emailing support@brea.ai. Checked 24 October 2026.",
      "pricing": {
        "amount": 0,
        "currency": "USD",
        "billing_period": null,
        "allowances": "Vendor homepage: 'Free, with every feature included'",
        "usage_charges": null,
        "verified_at": "2026-10-24",
        "source_url": "https://brea.ai",
        "note": "Brea's homepage (read 22 October 2026) says 'Free, with every feature included', and the standalone pricing page now redirects to the homepage. The earlier Free/Starter $9 per month tiers (verified 20 October 2026) are no longer published, so treat the change as current vendor wording, not a guaranteed permanent state.",
        "class": "free"
      },
      "sources": [
        "https://brea.ai",
        "https://brea.ai/privacy"
      ],
      "changelog": [
        {
          "date": "2026-10-20",
          "change": "Initial documentation research from homepage, pricing page and privacy policy. No hands-on testing."
        },
        {
          "date": "2026-10-22",
          "change": "Field-level audit: removed stale Free/Starter/Pro tier restrictions (pricing page gone; site advertises free with every feature). Dropped the image creation/editing capability, which no current Brea page documents. Privacy policy re-read (still 15 October 2026, claims unchanged: Google no-training, 90-day analytics deletion, account-deletion terms, App Store subscription caveat)."
        },
        {
          "date": "2026-10-24",
          "change": "Re-verified: 'Free, with every feature included' still on homepage. Privacy policy re-issued 23 October 2026 - new named advertising trackers (Meta Pixel/Conversions API, OpenAI Ads pixel/Conversions API receiving hashed email, name and billing location on signup/subscription), Apple/Google sign-in detail, encrypted saved logins, messaging delivery via SMS/iMessage/WhatsApp/Telegram. Integrations count now 2,200+."
        }
      ],
      "coverage_rows": [
        [
          "Where it runs",
          "Documented",
          "Website, web dashboard and iOS app; Apple processes in-app payments.",
          "https://brea.ai/privacy"
        ],
        [
          "Workflows and schedules",
          "Documented",
          "Examples include an 8am inbox summary and price-drop texts. Free tier allows 3 workflows.",
          "https://brea.ai"
        ],
        [
          "Browser automation",
          "Documented",
          "Vendor says Brea can fill forms, compare prices and book appointments on websites. Safety and success rates not tested.",
          "https://brea.ai"
        ],
        [
          "Cost",
          "Partially listed",
          "Free $0 and Starter $9/month confirmed; Pro price not readable on the rendered page.",
          "https://brea.ai/pricing"
        ],
        [
          "Training and retention",
          "Policy reviewed",
          "Google user data excluded from model training; conversations and files deleted on account deletion; vendor analytics copy deleted after 90 days.",
          "https://brea.ai/privacy"
        ]
      ],
      "setup_steps": [
        "Start on the Free tier with one integration and a single scheduled workflow.",
        "Verify the workflow output against your own inbox or account before trusting it.",
        "Review notification settings; reply previews can appear on the lock screen.",
        "Save site logins only for low-risk sites until its browser automation proves itself.",
        "Remember that account deletion does not cancel an App Store subscription."
      ],
      "price_tags": [
        "free"
      ],
      "price_summary": "Free, with every feature included (vendor homepage, Oct 2026)",
      "page_url": "https://agentcomparison.net/agents/brea/"
    },
    {
      "id": "shuffle",
      "name": "Shuffle",
      "vendor": "DNFT, Inc.",
      "kind": "product",
      "category": "AI social media manager",
      "group": "work-business",
      "summary": "Shuffle (DNFT, Inc.) documents an AI social media manager for small businesses: it makes posts (videos, photos, stories and carousels), publishes to connected accounts, and answers DMs, comments and Google reviews, with each post approved by text before it goes out. The earlier group-chat assistant survives as a free companion chat with an $8/month Pro tier.",
      "one_liner": "An AI social media manager for small businesses: it makes and runs your posts, and you approve by text.",
      "best_fit": "A small business, shop or creator that wants its social accounts run end to end, with final approval by text.",
      "poor_fit": "Anyone wanting a general personal assistant, annual-billing discounts, or a guarantee that nothing they type trains models.",
      "decision_note": "Shuffle's plans are now published: Starter $99, Business $199 and Agency $299 per month, monthly billing with a free 3-video start. Its privacy policy (updated 14 October 2026) says connected social and business account data is never used for training, sold or used for advertising, but the same policy allows training on other user content, so what you type outside connected accounts is treated differently. Publishing requires your text approval per the homepage.",
      "availability": "Live with monthly plans (first 3 videos free, no card) and sign-in on the site. Posting depends on the connected account; Instagram, TikTok, Facebook, YouTube and Google Business are listed. Reliability and output quality were not tested.",
      "last_checked": "2026-10-24",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Content creation: videos, photos, stories and carousels with ideas and captions",
          "source_url": "https://www.textshuffle.com/plans",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Publishing to connected social accounts after text approval",
          "source_url": "https://textshuffle.com",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Customer replies: DMs, comments and Google reviews from your business details and reply rules",
          "source_url": "https://textshuffle.com",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "MCP connection so your own agent (Claude, Codex) can create content and manage accounts",
          "source_url": "https://textshuffle.com",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Companion chat assistant (free tier; Pro $8/month)",
          "source_url": "https://www.textshuffle.com/plans",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Posts are approved by text before publishing, per the homepage ('You approve by text', 'Your final say').",
        "Customer replies (DMs, comments, Google reviews) follow business details and reply rules you set; you choose what to turn on and can step in anytime.",
        "Disconnecting a connected account deletes the settings, learned voice, facts, conversation records and links stored for it, per the privacy policy.",
        "Group-chat participants can remove the companion by texting STOP, per the terms."
      ],
      "limitations": [
        "Content counts on plans are a typical mix against one monthly balance; output varies with video length, carousel size and mix, per the plans page.",
        "Privacy policy (1.5) permits training AI models on user content other than connected social/business accounts and Google API data.",
        "Posting and reply availability depends on the connected account, per the vendor.",
        "Reliability, content quality and reply accuracy were not tested.",
        "The companion chat product's terms beyond 'free' and '$8/month Pro' are not published on the pages read."
      ],
      "runtime": [
        "Vendor-managed service; connects to Instagram, TikTok, Facebook, YouTube and Google Business; ads platform support listed (Meta, Google, TikTok, LinkedIn, Pinterest, X, OpenAI, Reddit Ads)"
      ],
      "privacy_security": "Privacy policy effective 8 September 2026, last updated 14 October 2026 (DNFT, Inc.). Connected social and business account data - including messages from people who contact you - is not used to train AI models, not sold and not used for advertising, and is deleted when you disconnect. Google API data gets the same never-train treatment. Separately, the policy says the service may train and improve AI models on other user content, excluding those two categories. Saved browsing sessions are held by the infrastructure provider that runs the browser. Payments via Stripe (token + last four only). Contact: privacy@dnftinc.com.",
      "pricing": {
        "amount": 99,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Free start: first 3 videos free, no card required",
        "usage_charges": "One monthly content balance per plan; typical mixes listed (Starter 10 reels + 20 stories; Business 22 + 30; Agency 30 + 30 + 12 carousels), swappable across formats, counts based on 15-second videos",
        "verified_at": "2026-10-24",
        "source_url": "https://www.textshuffle.com/plans",
        "note": "Starter $99/month, Business $199/month, Agency $299/month, monthly billing, cancel anytime. Companion chat: free tier, Shuffle Pro $8/month.",
        "class": "amount"
      },
      "sources": [
        "https://textshuffle.com",
        "https://www.textshuffle.com/plans",
        "https://www.textshuffle.com/privacy-policy",
        "https://www.textshuffle.com/terms"
      ],
      "changelog": [
        {
          "date": "2026-10-21",
          "change": "Initial documentation research from homepage. No pricing or privacy page found. No hands-on testing."
        },
        {
          "date": "2026-10-24",
          "change": "Full rewrite: vendor pivoted from an iMessage/SMS group-chat assistant to an AI social media manager for small businesses. Plans page now published ($99-$299/month, free start); privacy policy (added 8 Sep 2026, updated 14 Oct 2026) and terms now exist. Old product survives as companion chat (free, Pro $8/month). Record moved to the work-business group."
        }
      ],
      "coverage_rows": [
        [
          "Content creation",
          "Documented",
          "Videos, photos, stories and carousels with ideas and captions, made for your business and voice.",
          "https://www.textshuffle.com/plans"
        ],
        [
          "Publishing and approvals",
          "Documented",
          "You approve each post by text before it goes out; publish to multiple accounts where supported.",
          "https://textshuffle.com"
        ],
        [
          "Customer replies",
          "Documented",
          "Answers DMs, comments and Google reviews from your business details and reply rules; you choose what to turn on.",
          "https://textshuffle.com"
        ],
        [
          "Cost",
          "Documented",
          "Free 3-video start; Starter $99, Business $199, Agency $299 per month, monthly billing. Companion chat free, Pro $8/month.",
          "https://www.textshuffle.com/plans"
        ],
        [
          "Data use",
          "Documented with a caveat",
          "Connected-account and Google data never used for training; other user content may be used to train models per the privacy policy.",
          "https://www.textshuffle.com/privacy-policy"
        ]
      ],
      "setup_steps": [
        "Start with the free 3 videos on one account before choosing a plan.",
        "Set your business details and reply rules before turning on DM, comment or review replies.",
        "Check the privacy policy distinction: connected-account data is never trained on, other chat content may be.",
        "Confirm your platforms are supported (Instagram, TikTok, Facebook, YouTube, Google Business listed) before paying."
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Free 3-video start; plans $99-$299/month; companion chat free, Pro $8/month",
      "page_url": "https://agentcomparison.net/agents/shuffle/"
    },
    {
      "id": "kicker",
      "name": "Kicker",
      "vendor": "Companion Intelligence Inc.",
      "kind": "product",
      "category": "personal health agent",
      "group": "personal-assistants",
      "summary": "Kicker (Companion Intelligence Inc.) documents a personal health agent that picks your three daily priorities, blocks time at your gym, shifts meetings after a short night, and connects Oura, WHOOP and Fitbit alongside email, calendars and notes. Accounts are free to create; the vendor claims 150,000+ people use it.",
      "one_liner": "A personal health agent that picks your daily priorities and connects your wearables and apps.",
      "best_fit": "Someone who wants a proactive health-and-habits agent tied to their wearables, calendar and everyday apps.",
      "poor_fit": "Anyone needing published pricing, or a privacy policy that matches the current product before connecting health data.",
      "decision_note": "Kicker has launched since our last check: the homepage no longer shows a waitlist and the FAQ confirms free account creation with sign-in by phone number or Google. Paid plans are only shown before you subscribe, so cost is still unpublished. Its privacy policy (updated 20 September 2026) still describes the earlier voice-agent platform for calls to businesses, not the current health agent, so data practices for the features on the homepage are only partly covered.",
      "availability": "Live per the homepage and FAQ (read 24 October 2026): create a free account and sign in with your phone number or Google, then connect devices and apps or skip and add them later in Workspace. 'Trusted by 150,000+ people' is a vendor claim. Reliability was not tested.",
      "last_checked": "2026-10-24",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Proactive daily priority selection (three things that matter today)",
          "source_url": "https://usekicker.com",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Calendar management: blocks gym time, shifts meetings after a short night",
          "source_url": "https://usekicker.com",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Wearable and app connections: Oura, WHOOP, Fitbit plus email, calendars and notes",
          "source_url": "https://usekicker.com/faq",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "You choose which devices and apps to connect and can remove them, per the FAQ.",
        "Account, connections and history can be deleted from Preferences in Workspace, per the FAQ.",
        "Optional model-training opt-in for anonymised voice data exists per the privacy policy and excludes Google User Data."
      ],
      "limitations": [
        "Paid plan prices are not published; the FAQ says they are shown before you subscribe.",
        "The privacy policy (updated 20 September 2026) still describes a voice agent platform for calls to businesses, not the current health agent.",
        "'Trusted by 150,000+ people' is a vendor marketing claim.",
        "Proactive examples (picking priorities, blocking gym time, shifting meetings) are vendor descriptions, not tested behaviour."
      ],
      "runtime": [
        "Vendor-managed app and Workspace; connects Oura, WHOOP, Fitbit plus email, calendars, notes and other everyday apps"
      ],
      "privacy_security": "Privacy policy effective 20 April 2026, last updated 20 September 2026 (Companion Intelligence Inc.). It still describes a voice agent platform for calls to businesses, so coverage of the current health-agent features is partial. It collects voice recordings and call transcripts, offers an optional model-training opt-in that excludes Google User Data, and states Google User Data is never sold, used for advertising or used to train generalised AI models.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": "FAQ: create an account and connect devices and apps for free",
        "usage_charges": null,
        "verified_at": "2026-10-24",
        "source_url": "https://usekicker.com/faq",
        "note": "Free account creation per the FAQ (24 October 2026); paid plans are shown before you subscribe but not published on the site.",
        "class": "see_vendor"
      },
      "sources": [
        "https://usekicker.com",
        "https://usekicker.com/privacy",
        "https://usekicker.com/faq"
      ],
      "changelog": [
        {
          "date": "2026-10-21",
          "change": "Initial documentation research from homepage and privacy policy. Product is pre-launch; no hands-on testing possible."
        },
        {
          "date": "2026-10-24",
          "change": "Rewrite: Kicker has launched - homepage no longer shows the waitlist and the FAQ confirms free accounts (phone or Google sign-in), wearable connections (Oura, WHOOP, Fitbit) and Workspace data controls. Privacy policy still describes the earlier voice-agent platform."
        }
      ],
      "coverage_rows": [
        [
          "Availability",
          "Documented",
          "Live: free account, sign in with phone number or Google. \"Trusted by 150,000+ people\" is a vendor claim.",
          "https://usekicker.com/faq"
        ],
        [
          "Proactive behaviour",
          "Vendor examples",
          "Picks three daily priorities, blocks gym time, shifts meetings after a short night. Not tested.",
          "https://usekicker.com"
        ],
        [
          "Connections",
          "Documented",
          "Oura, WHOOP, Fitbit plus email, calendars, notes and everyday apps, managed in Workspace.",
          "https://usekicker.com/faq"
        ],
        [
          "Cost",
          "Documented",
          "Free account creation; paid plans shown before you subscribe, not published.",
          "https://usekicker.com/faq"
        ],
        [
          "Data controls",
          "Documented",
          "Delete account, connections and history from Workspace Preferences.",
          "https://usekicker.com/faq"
        ]
      ],
      "setup_steps": [
        "Create a free account without connecting health devices first, and see what it asks for.",
        "Read the privacy policy knowing it still describes the earlier voice-call platform.",
        "Check the price of any paid plan before subscribing, since none are published."
      ],
      "price_tags": [
        "free"
      ],
      "price_summary": "Free account; paid plans not published",
      "page_url": "https://agentcomparison.net/agents/kicker/"
    },
    {
      "id": "tomo",
      "name": "Tomo",
      "vendor": "Mapo Labs, Inc.",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Tomo (Mapo Labs, Inc.) documents a personal AI that lives in your texts: it learns your goals, joins group chats, sends reminders, searches the web, manages calendar and email, and messages you proactively.",
      "one_liner": "A text-based personal AI for goals, accountability, reminders and life admin.",
      "best_fit": "A text-based accountability and life-organisation assistant that messages you first.",
      "poor_fit": "Users who want a fixed published price ladder: only the base plan is priced publicly and prices vary by user.",
      "decision_note": "Tomo now publishes a starting price: the base plan is from $19.99/month with pricier plans for higher messaging limits, but the FAQ says pricing may vary by user (promotions), so confirm your actual price at signup. Its privacy policy (June 2026) excludes Google Workspace and Apple Health data from AI training; broader model-training practices were not established in the sections reviewed.",
      "availability": "Vendor says to text your Tomo; login and get-started links are on the site. Claims '500,000+ people' use it. Signup and reliability were not tested.",
      "last_checked": "2026-10-24",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Two-way texting with a personal AI",
          "source_url": "https://tomo.ai/about",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Group chats",
          "source_url": "https://tomo.ai/about",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Reminders and proactive messages",
          "source_url": "https://tomo.ai/about",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Calendar and email management",
          "source_url": "https://tomo.ai/about",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Web search, video watching and photo editing",
          "source_url": "https://tomo.ai/about",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "low",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Apple Health connection (activity, sleep, heart rate) with explicit consent, per the privacy policy",
          "source_url": "https://tomo.ai/privacy",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Connects to calendar and email per the about page; connection scopes and approval mechanics were not established.",
        "Proactive outbound messages are part of the product; frequency controls were not established."
      ],
      "limitations": [
        "Only the base plan's $19.99/month start is published; the full plan ladder is not, and prices vary by user (promotions).",
        "The '500,000+ people' figure is a vendor claim.",
        "General model-training and retention specifics were not established in the sections reviewed.",
        "Reliability and response quality were not tested."
      ],
      "runtime": [
        "Vendor-managed assistant reached by text"
      ],
      "privacy_security": "Privacy policy last updated 23 June 2026 (Mapo Labs, Inc.). Personal data is retained only as long as necessary for the purposes described, then deleted or anonymised. Google Workspace data is not used to develop, improve or train generalised AI/ML models. Apple HealthKit data (optional, with consent) powers the agent; it is not sold, not used for advertising and not used to train generalised AI models. Advertising/targeting cookies (Meta, TikTok) are used on marketing pages but not on the privacy or terms pages; California residents can opt out of this sharing (CCPA 'Do Not Share') by email. Messages, reminders and memories can be cleared from dashboard support; complete deletion via a data deletion form, processed within 30 days.",
      "pricing": {
        "amount": 19.99,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Base plan; other plans with more expansive messaging limits, per the FAQ",
        "usage_charges": null,
        "verified_at": "2026-10-24",
        "source_url": "https://tomo.ai/about",
        "note": "FAQ (read 24 October 2026): the base plan starts at $19.99/month, other plans have more expansive messaging limits, cancel anytime in the dashboard. Pricing may vary by user and is subject to change; promotions and discounts mean not everyone gets the same price. Terms: subscriptions auto-renew, non-refundable except as required by law. Payments via Stripe.",
        "class": "amount"
      },
      "sources": [
        "https://tomo.ai",
        "https://tomo.ai/about",
        "https://tomo.ai/privacy",
        "https://tomo.ai/terms"
      ],
      "changelog": [
        {
          "date": "2026-10-21",
          "change": "Initial documentation research from homepage, about page, privacy policy and terms. No hands-on testing."
        },
        {
          "date": "2026-10-24",
          "change": "Pricing now published in the FAQ: base plan from $19.99/month, other plans with higher messaging limits, cancel in dashboard; prices vary by user (promotions). Privacy additions captured: Apple HealthKit section (health data powers the agent; never sold, used for ads or model training), CCPA Do-Not-Share opt-out, 30-day deletion form. Terms re-read (auto-renew, non-refundable except as required by law)."
        }
      ],
      "coverage_rows": [
        [
          "Messaging channels",
          "Documented",
          "Lives in your texts; group chats supported per the about page.",
          "https://tomo.ai/about"
        ],
        [
          "Proactive behaviour",
          "Documented",
          "Proactively messages you to keep you on track; frequency controls not established.",
          "https://tomo.ai/about"
        ],
        [
          "Calendar and email",
          "Documented",
          "Connects to calendar and email; scopes not established.",
          "https://tomo.ai/about"
        ],
        [
          "Cost",
          "Documented",
          "FAQ: base plan from $19.99/month, other plans with higher messaging limits; prices vary by user (promotions). Terms: auto-renewing, non-refundable except as required by law.",
          "https://tomo.ai/about"
        ],
        [
          "Training and retention",
          "Policy reviewed",
          "Google Workspace data excluded from AI training; personal data retained only as needed, then deleted or anonymised.",
          "https://tomo.ai/privacy"
        ]
      ],
      "setup_steps": [
        "Text Tomo and test a low-stakes request before connecting accounts.",
        "Confirm your actual price at signup; the FAQ says pricing may vary by user (promotions and discounts).",
        "Check how to control proactive message frequency before relying on it.",
        "Connect calendar and email only after reviewing what access is requested."
      ],
      "price_tags": [
        "listed"
      ],
      "price_summary": "From $19.99/month (base plan); prices vary by user",
      "page_url": "https://agentcomparison.net/agents/tomo/"
    },
    {
      "id": "asaply",
      "name": "asaply",
      "vendor": "Asaply Inc. (Delaware corporation)",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "asaply documents a text-first errand assistant: you text a number and it orders coffee and lunch ahead, books rides, handles tickets and reservations, runs group errands with split payments, and queues Apple Wallet passes - no app or account required, your phone number is your account.",
      "one_liner": "A number you text to get real-world errands done - coffee, rides, tickets - with payment handled in the thread.",
      "best_fit": "Getting small real-world errands handled by text - coffee and lunch pickups, rides, tickets - without installing an app.",
      "poor_fit": "People outside its early-access cities, or anyone wanting a flat subscription instead of per-order fees.",
      "decision_note": "asaply is priced per order (your total plus a small service fee, no subscription) and is live in early access with coverage rolling out neighborhood by neighborhood, so availability depends on your city. Its legal page (updated 11 October 2026) is unusually detailed about approvals and Google data handling.",
      "availability": "Live in early access per its FAQ; rolling out neighborhood by neighborhood and coverage varies by city. The vendor does not publish a service-area list (read 21 October 2026).",
      "last_checked": "2026-10-24",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Order-ahead pickups by text (coffee, lunch, local merchants)",
          "source_url": "https://asaply.ai/",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Ride booking with timing advice",
          "source_url": "https://asaply.ai/",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Tickets and reservations (events, museums, transit passes)",
          "source_url": "https://asaply.ai/",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Group errands with split payments",
          "source_url": "https://asaply.ai/",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Recurring routine orders (\"nudges\")",
          "source_url": "https://asaply.ai/",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Apple Wallet passes and flight check-in",
          "source_url": "https://asaply.ai/",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Google Calendar read and approved changes",
          "source_url": "https://asaply.ai/legal",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Gmail read, alerts and approved replies",
          "source_url": "https://asaply.ai/legal",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Location-aware suggestions (optional location sharing)",
          "source_url": "https://asaply.ai/legal",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Google Calendar read and changes only after connecting via Google sign-in; event changes beyond simple adds require exact-change approval, with 30-minute undo.",
        "Gmail read for questions and alerts; sending only after you approve the exact draft; it never sends email on its own, per the legal page.",
        "Group chats can see only your free/busy blocks, never event details, per the legal page."
      ],
      "limitations": [
        "Availability varies by city during the early-access rollout; no coverage list published.",
        "Total cost is per order plus a service fee that is only shown before confirming each order; no public fee schedule.",
        "Reliability, merchant coverage and timing were not tested."
      ],
      "runtime": [
        "Vendor-managed assistant reached by texting its number (SMS/iMessage)",
        "Optional app mentioned in the legal page, not required"
      ],
      "privacy_security": "Privacy policy and terms published at asaply.ai/legal (Asaply Inc., updated 11 October 2026): no sale of personal data, encryption in transit and at rest, Google API Limited Use compliance, AI processing via Google Cloud Vertex AI without training on user data, location history auto-deleted within 90 days, and disconnect/delete by texting \"disconnect google\".",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": "Per order: you pay the order total plus a small service fee, shown before confirming; payment in-thread with Apple Pay.",
        "verified_at": "2026-10-24",
        "source_url": "https://asaply.ai/",
        "note": "FAQ (re-read 24 October 2026): pay for your order plus a small service fee, no subscription, total always shown before confirming; payment in-thread with Apple Pay. Terms add that fees are non-refundable except where required by law.",
        "class": "paid"
      },
      "sources": [
        "https://asaply.ai/",
        "https://asaply.ai/legal"
      ],
      "changelog": [
        {
          "date": "2026-10-21",
          "change": "Initial documentation research from homepage, FAQ and legal page. No hands-on testing."
        },
        {
          "date": "2026-10-24",
          "change": "Re-verified homepage, FAQ and legal page (still updated 11 October 2026): per-order pricing, early-access rollout, Apple Pay, Google and location handling all unchanged. Terms fee rule (non-refundable except where required by law) added to the pricing note."
        }
      ],
      "coverage_rows": [
        [
          "Messaging channels",
          "Documented",
          "Text a dedicated number (SMS/iMessage); no app or account needed. Group errands in group chats.",
          "https://asaply.ai/"
        ],
        [
          "Errands and ordering",
          "Documented",
          "Coffee/lunch pickups, rides, tickets, reservations, wallet passes; not tested.",
          "https://asaply.ai/"
        ],
        [
          "Connected accounts",
          "Documented with approvals",
          "Optional Google Calendar and Gmail via Google sign-in; sends and sensitive changes need your approval.",
          "https://asaply.ai/legal"
        ],
        [
          "Cost",
          "Documented structure",
          "Per-order total plus small service fee; no subscription; fee schedule not published.",
          "https://asaply.ai/"
        ],
        [
          "Privacy",
          "Documented",
          "Detailed policy; no data sale; Google Limited Use; Vertex AI processing without training.",
          "https://asaply.ai/legal"
        ],
        [
          "Availability",
          "Limited rollout",
          "Early access, neighborhood-by-neighborhood; coverage varies by city.",
          "https://asaply.ai/"
        ]
      ],
      "setup_steps": [
        "Text the asaply number and answer its three onboarding questions.",
        "Confirm it covers your neighborhood before relying on it.",
        "Check the total (order + service fee) before confirming each order."
      ],
      "price_tags": [
        "usage"
      ],
      "price_summary": "Per-order service fee, no subscription (per FAQ)",
      "page_url": "https://agentcomparison.net/agents/asaply/"
    },
    {
      "id": "orchid",
      "name": "Orchid",
      "vendor": "Orchid",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Orchid documents a personal assistant that lives in iMessage: it reads your connected Gmail, drafts replies in your voice, schedules on your Google Calendar with your approval, preps you before meetings, plans travel, and runs recurring habits - all from a text thread, with no app to install. The vendor has also launched Keiki, a platform for building customer-facing agents across iMessage, SMS, WhatsApp, Slack, Telegram and email.",
      "one_liner": "A personal assistant in your iMessage thread: inbox triage, drafts in your voice, scheduling and meeting prep, paid from $29.99/month.",
      "best_fit": "Founders, lawyers and other busy professionals who want inbox triage, drafting and scheduling handled from an iMessage thread.",
      "poor_fit": "Anyone wanting a free tier, or needing integrations beyond Gmail and Google Calendar today.",
      "decision_note": "Orchid is paid from $29.99/month with no free tier (the vendor argues an assistant that reads your inbox should not be paid for with your data). Only Gmail and Google Calendar connect today; Outlook, Slack, Notion and Linear are listed as on the way. Approval-first: drafts stay drafts and calendar holds need your sign-off.",
      "availability": "Live per its site; setup starts by texting +1 (415) 605-3216 (the number linked from its homepage, read 24 October 2026) and connecting Gmail and Google Calendar via Google sign-in.",
      "last_checked": "2026-10-24",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Inbox triage and reply drafts in your voice (Gmail)",
          "source_url": "https://orchid.ai/blog/orchid-ai-assistant-imessage",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Scheduling and calendar holds with approval (Google Calendar)",
          "source_url": "https://orchid.ai/blog/orchid-ai-assistant-imessage",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Meeting prep briefings by text",
          "source_url": "https://orchid.ai/blog/orchid-ai-assistant-imessage",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Conditional reminders",
          "source_url": "https://orchid.ai/blog/orchid-ai-assistant-imessage",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Voice notes in the thread",
          "source_url": "https://orchid.ai/blog/orchid-ai-assistant-imessage",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Recurring habits (news digests, trackers, check-ins)",
          "source_url": "https://orchid.ai/",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Travel planning and itineraries",
          "source_url": "https://orchid.ai/faq",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Gift sourcing and personal admin",
          "source_url": "https://orchid.ai/faq",
          "observed_at": "2026-10-21",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Gmail and Google Calendar connect via Google sign-in (OAuth); access revocable from the Google account at any time.",
        "Email replies stay as drafts until you approve; calendar holds and bookings happen only with your approval, per the FAQ."
      ],
      "limitations": [
        "No free tier: paid from $29.99/month (vendor blog, read 24 October 2026).",
        "Only Gmail and Google Calendar connect today; Outlook, Slack, Notion and Linear are not yet available.",
        "Draft quality, voice matching and reliability were not tested.",
        "The vendor now also sells Keiki, a separate business platform for building customer-facing agents; it is not part of the personal assistant record here."
      ],
      "runtime": [
        "Vendor-managed assistant inside an iMessage thread",
        "Web app at app.orchid.ai for connecting accounts and settings, not needed day to day"
      ],
      "privacy_security": "Privacy policy published (updated 16 January 2026): OAuth 2.0 least-privilege integrations, data deletion on request, approval/edit before actions execute, audit trails. The FAQ states Orchid never sells your data, shares it, or uses it to train any model.",
      "pricing": {
        "amount": 29.99,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Flat monthly price, no meter on messages, per the vendor pricing blog.",
        "usage_charges": null,
        "verified_at": "2026-10-24",
        "source_url": "https://orchid.ai/blog/orchid-ai-assistant-imessage",
        "note": "Vendor blog (read 21 October 2026): paid, from $29.99/mo flat, no free tier; student and founder deals exist.",
        "class": "amount"
      },
      "sources": [
        "https://orchid.ai/",
        "https://orchid.ai/faq",
        "https://orchid.ai/blog/orchid-ai-assistant-imessage",
        "https://orchid.ai/privacy",
        "https://orchid.ai/blog/introducing-keiki"
      ],
      "changelog": [
        {
          "date": "2026-10-21",
          "change": "Initial documentation research from homepage, FAQ, iMessage setup blog and privacy policy. No hands-on testing."
        },
        {
          "date": "2026-10-24",
          "change": "Setup number updated to +1 (415) 605-3216 (current homepage sms link). New Keiki platform launch noted (customer-facing agent builder across messaging channels). Pricing re-verified from the vendor blog ($29.99/month, no free tier); FAQ re-read (Gmail and Google Calendar today; Outlook, Slack, Notion and Linear still on the way); privacy policy unchanged (16 January 2026)."
        }
      ],
      "coverage_rows": [
        [
          "Messaging channels",
          "Documented",
          "iMessage thread (blue bubble) with a dedicated number; voice notes supported.",
          "https://orchid.ai/blog/orchid-ai-assistant-imessage"
        ],
        [
          "Email and calendar",
          "Documented with approvals",
          "Gmail triage and drafts, Google Calendar holds; nothing sends or books without your approval.",
          "https://orchid.ai/faq"
        ],
        [
          "Other integrations",
          "Announced, not live",
          "Outlook, Slack, Notion and Linear listed as on the way.",
          "https://orchid.ai/faq"
        ],
        [
          "Cost",
          "Documented",
          "Paid from $29.99/month, flat, no free tier; student and founder deals.",
          "https://orchid.ai/blog/orchid-ai-assistant-imessage"
        ],
        [
          "Privacy",
          "Documented",
          "Policy published; OAuth least-privilege; no sale, sharing or training per FAQ.",
          "https://orchid.ai/privacy"
        ]
      ],
      "setup_steps": [
        "Text +1 (415) 605-3216 to start the thread.",
        "Connect Gmail and Google Calendar via Google sign-in from the link Orchid sends.",
        "Set approval preferences before letting it draft or schedule for real contacts."
      ],
      "price_tags": [
        "listed"
      ],
      "price_summary": "From $29.99/month, no free tier (vendor blog)",
      "page_url": "https://agentcomparison.net/agents/orchid/"
    },
    {
      "id": "grok-bot",
      "name": "Grok Bot",
      "vendor": "xAI",
      "kind": "product",
      "category": "work assistant",
      "group": "work-business",
      "summary": "Grok Bot is xAI's desktop and iOS product for creating AI teammates: you give a bot a task like you would a colleague, it signs into your apps and websites with your logins, works through the job, and comes back when your approval is needed. Multiple bots run in parallel.",
      "one_liner": "xAI's bot teammates that sign into your tools and return finished work.",
      "best_fit": "Teams and solo operators who want an agent to run multi-step work across their existing logged-in tools.",
      "poor_fit": "Anyone on Android or web-only workflows, or who cannot grant an agent access to their signed-in accounts.",
      "decision_note": "Grok Bot works through your own logged-in sessions on your apps (the launch page shows a bot asking you to sign in to Salesforce so it can proceed), which is a broad access grant to weigh. Approval checkpoints exist (drafts held for your ok). Pricing is public: Premium $30/month, Super $100/month, Ultra $200/month, all billed monthly with weekly usage included and token-cost overage.",
      "availability": "Live per its site: direct downloads for Linux (.deb, AppImage, .rpm; x64 and arm64), macOS (Apple silicon and Intel) and Windows, plus iOS and Android apps (verified in a live browser 24 October 2026). Grok Bot is included with Premium, Super, Ultra, Teams and Enterprise plans; Enterprise setup via contact sales.",
      "last_checked": "2026-10-24",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Task delegation to parallel AI 'teammates'",
          "source_url": "https://x.ai/bot",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Signs into your apps and websites and operates them (e.g. Salesforce, Gmail, LinkedIn shown)",
          "source_url": "https://x.ai/bot",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Approval checkpoints: drafts and actions held for your ok",
          "source_url": "https://x.ai/bot",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Persistent context: bots keep how you work and improve over time (vendor claim)",
          "source_url": "https://x.ai/bot",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Routines: watch you complete a workflow once, then run it on a schedule",
          "source_url": "https://x.ai/bot",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Multi-bot threads: bots pass work between themselves on a shared cloud computer",
          "source_url": "https://x.ai/bot",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "You sign the bot into each tool yourself (the page shows a per-app sign-in step, e.g. Salesforce).",
        "Bots return for approval at defined checkpoints, per the launch page (e.g. drafts held for your ok)."
      ],
      "limitations": [
        "Usage beyond the weekly included amount is billed at token cost, so heavy use costs more than the sticker price.",
        "Bots share one persistent cloud computer per user (files, browser, logins); isolation is per user, not per bot.",
        "Bots act inside your own logged-in sessions; the blast radius of a mistake is your own account access.",
        "Reliability and task success rates were not tested."
      ],
      "runtime": [
        "Vendor-managed bots in the Grok Bot app: Linux, macOS and Windows desktop downloads plus iOS and Android apps (all platforms offered on the page, verified 24 October 2026)"
      ],
      "privacy_security": "Per its page, Grok Bot uses the same SSO, auth and privacy mode as Cursor; the shared cloud computer is encrypted in transit and at rest, with training opt-out, and sensitive actions can go through Auto Review. Enterprise admins can set DLP, certs, proxies and network controls. These are vendor claims; review xAI's terms before connecting work accounts.",
      "pricing": {
        "amount": 30,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Premium $30/month billed monthly (weekly Grok Bot usage included); Super $100/month (3.5x Premium usage, priority model access at peak times); Ultra $200/month (12x Premium usage, earliest access to all products and models). Also included with Teams and Enterprise plans.",
        "usage_charges": "Additional usage beyond the weekly included amount is billed based on token cost (per the page's pricing FAQ)",
        "verified_at": "2026-10-24",
        "source_url": "https://x.ai/bot",
        "note": "x.ai/bot (verified in a live browser 22 October 2026) lists Premium at $30/month, Super at $100/month and Ultra at $200/month, all billed monthly, with weekly Grok Bot usage included and extra usage billed at token cost. Grok Bot is also included with eligible existing Premium, Super, Ultra, Teams or Enterprise plans; Teams and Enterprise access is via contact sales.",
        "class": "amount"
      },
      "sources": [
        "https://x.ai/bot"
      ],
      "changelog": [
        {
          "date": "2026-10-22",
          "change": "Initial documentation research from the Grok Bot launch page. No hands-on testing."
        },
        {
          "date": "2026-10-22",
          "change": "Pricing verified in a live browser on x.ai/bot: Premium $30, Super $100 and Ultra $200 per month, billed monthly, with weekly usage included and token-cost overage; platform availability and privacy FAQ updated from the same page."
        },
        {
          "date": "2026-10-24",
          "change": "Re-verified in a live browser: Premium $30, Super $100 and Ultra $200 per month billed monthly with weekly usage and token-cost overage all unchanged. Now included with Teams and Enterprise plans. Downloads now cover Linux (x64/arm64), macOS (Apple silicon and Intel), Windows, iOS and Android. New documented features: routines (watch once, run on a schedule) and multi-bot threads; FAQ confirms bots share one persistent cloud computer per user."
        }
      ],
      "price_tags": [
        "listed"
      ],
      "price_summary": "Premium $30, Super $100, Ultra $200 per month (x.ai/bot, Oct 2026)",
      "page_url": "https://agentcomparison.net/agents/grok-bot/"
    },
    {
      "id": "zapia",
      "name": "Zapia",
      "vendor": "Zapia",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Zapia (BrainLogic AI) pitches itself as 'the AI that runs your WhatsApp' for Latin America: from its own iOS/Android app and web app it connects to WhatsApp, Gmail, Google Calendar, Google Drive, Outlook, Slack and iFood to summarise chats, send and schedule messages, manage your calendar, draft emails, shop and make bookings. Free tier plus Pro ($9.99/month) and Ultra ($99.99/month) plans, both with 7-day free trials.",
      "one_liner": "A personal AI for Latin America that runs your WhatsApp and connected apps from its own app.",
      "best_fit": "Spanish- and Portuguese-speaking users who want a proactive assistant app for email, scheduling and life admin.",
      "poor_fit": "Anyone wanting the assistant as a contact inside WhatsApp or iMessage rather than a separate app.",
      "decision_note": "Zapia's FAQ still says it lives in its own app and on the web, not as a WhatsApp contact - it connects to your WhatsApp account to organise and send messages, so it does not meet our messaging-table inclusion rule and is directory-only. Plans are now published on the homepage: Free $0, Pro $9.99/month and Ultra $99.99/month with 7-day trials. Aimed at Latin American users; its privacy policy names the third-party AI providers that process user content (OpenAI, Anthropic, Google, Together AI, Tencent US, Alibaba US) and discloses affiliate compensation on product recommendations.",
      "availability": "Live per its site; App Store and Google Play downloads offered, plus web login. Aimed at Latin America (iFood connector, Brazil DPO, Spanish and Portuguese). Claims 4.8 stars on both stores - a vendor claim (read 24 October 2026).",
      "last_checked": "2026-10-24",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Proactive monitoring of deadlines, email and invoices with summaries",
          "source_url": "https://zapia.com/",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Drafting email replies in the background",
          "source_url": "https://zapia.com/",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Rescheduling your day when meetings move",
          "source_url": "https://zapia.com/",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Document review on request (e.g. contracts before signing)",
          "source_url": "https://zapia.com/",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "WhatsApp connector: organises chats, summarises conversations, sends and schedules messages",
          "source_url": "https://zapia.com/",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Connectors: Gmail, Google Calendar, Google Drive, Outlook, Slack, iFood, plus custom connectors",
          "source_url": "https://zapia.com/",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Monitors deadlines, email and invoices; drafts replies and rearranges your day when conflicts appear.",
        "Transcribes voice notes when you connect your account.",
        "Handles data of third parties you involve (contacts, message recipients, calendar invitees) - flagged in its own policy."
      ],
      "limitations": [
        "Not a messaging-service contact: it runs in Zapia's own app and web app and connects to your WhatsApp account.",
        "User content is processed by named third-party AI providers (OpenAI, Anthropic, Google, Together AI, Tencent US, Alibaba US) per the privacy policy.",
        "Zapia may receive affiliate compensation when users buy recommended products, per its privacy policy.",
        "The '4.8★ on the App Store & Google Play' figure is a vendor claim.",
        "Task reliability and response quality were not tested."
      ],
      "runtime": [
        "Vendor-managed assistant in the Zapia iOS/Android apps and web app"
      ],
      "privacy_security": "Privacy policy last updated 5 October 2026 (Brainlogic AI SAS). It does not sell or share personal data with third parties for advertising or marketing. User content - messages, uploads, voice notes, calendar data - is processed by named third-party AI providers (OpenAI, Anthropic, Google, Together AI, Tencent hosted in the US, Alibaba hosted in the US) to provide the service; Zapia says it does not sell personal data to any AI provider. The homepage FAQ adds that models are not trained on your data by default, data is encrypted in transit and at rest, and Zapia passes Google's CASA Tier 2 audit. A transparency section discloses affiliate compensation on product recommendations (anonymous transaction identifiers, products, values, conversion data shared with affiliate programmes). Retention is 'as long as necessary' for the described purposes; rights requests go through the app, with a Brazil DPO contact for LGPD.",
      "pricing": {
        "amount": 9.99,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Free $0: basic queries, app connectors, up to 20 scheduled tasks, watermarked images. Pro: advanced intelligence, up to 50 scheduled tasks, watermark-free images with the best models, priority support. Ultra $99.99/month: everything in Pro, 10x advanced intelligence, up to 100 scheduled tasks, early access to new features.",
        "usage_charges": null,
        "verified_at": "2026-10-24",
        "source_url": "https://zapia.com/",
        "note": "Homepage plans section (read 24 October 2026): Free $0/month, Pro $9.99/month and Ultra $99.99/month, Pro and Ultra with a 7-day free trial; no commitment, cancel anytime. Prices shown in US dollars on the English site.",
        "class": "free_tier"
      },
      "sources": [
        "https://zapia.com/",
        "https://zapia.com/privacy"
      ],
      "changelog": [
        {
          "date": "2026-10-22",
          "change": "Initial documentation research. Directory-only: FAQ says it lives in its own app and web, not as a WhatsApp contact. No hands-on testing."
        },
        {
          "date": "2026-10-24",
          "change": "Plans now published on the homepage: Free $0, Pro $9.99/month, Ultra $99.99/month with 7-day trials. Repositioned as \"the AI that runs your WhatsApp\" for Latin America with connectors (WhatsApp, Gmail, Calendar, Drive, Outlook, Slack, iFood, custom). Privacy policy (5 October 2026) names third-party AI providers processing user content and discloses affiliate compensation on recommendations; FAQ adds no-training-by-default, encryption and CASA Tier 2 audit."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Documented",
          "Homepage plans: Free $0 (20 scheduled tasks), Pro $9.99/month (50 tasks, advanced intelligence), Ultra $99.99/month (100 tasks, 10x intelligence); 7-day trials, cancel anytime.",
          "https://zapia.com/"
        ],
        [
          "Channel",
          "Own app + web",
          "Zapia runs in its own app and on the web, not as a WhatsApp contact; it can connect to your account to transcribe voice notes.",
          "https://zapia.com/"
        ],
        [
          "Sale of data",
          "Not sold",
          "Privacy policy: 'we do not sell or share personal data with third parties' for their own purposes; service providers and connected-app data are processed to run the service.",
          "https://zapia.com/privacy"
        ],
        [
          "Sensitive data",
          "Discouraged",
          "The policy asks users not to enter, share or submit sensitive personal information through the services unless required.",
          "https://zapia.com/privacy"
        ]
      ],
      "setup_steps": [
        "Download the app (App Store or Google Play) or use the web - Zapia is not a WhatsApp contact.",
        "Start on the Free plan; Pro and Ultra have 7-day free trials.",
        "Review the privacy policy's third-party AI provider list before connecting accounts.",
        "Connect WhatsApp and other accounts only after checking what each connector can read and send."
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Free tier; Pro $9.99/month and Ultra $99.99/month, 7-day trials",
      "page_url": "https://agentcomparison.net/agents/zapia/"
    },
    {
      "id": "del",
      "name": "Del",
      "vendor": "Human Interface, Inc.",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Del is a personal assistant you reach by ordinary text message - no app, no login. It handles calendar briefings, email work, refunds and subscription cancellations, spending tracking, invoice follow-ups, marketplace listings, flight and hotel monitoring, insurance questions and rental listing watches. It remembers what you tell it.",
      "one_liner": "A personal helper you text for calendars, email, refunds, cancellations and everyday admin.",
      "best_fit": "Busy people who want to offload everyday admin - refunds, cancellations, briefings, monitoring - to a text thread with memory.",
      "poor_fit": "Anyone needing a published price before committing, or enterprise controls.",
      "decision_note": "Del's pitch is breadth: everyday admin by text with memory. It advertises a free two-week start, but no paid price is published on the pages we read, so long-term cost is unknown. Its privacy section states your data is never used to train AI and never sold, with encrypted passwords and deletion on account closure - vendor statements, not audited.",
      "availability": "Live per its site; you start by texting Del, no app install (read 22 October 2026).",
      "last_checked": "2026-10-24",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Morning calendar briefings by text",
          "source_url": "https://withdel.com/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Refund claims handled end to end (user testimonial describes a completed refund)",
          "source_url": "https://withdel.com/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Subscription cancellation on request",
          "source_url": "https://withdel.com/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Spending tracking and budgeting help",
          "source_url": "https://withdel.com/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Monitoring: flights, hotels, rental listings",
          "source_url": "https://withdel.com/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Memory of what you tell it (vendor claim)",
          "source_url": "https://withdel.com/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Works over ordinary text message - no app, no login: morning calendar briefings, email work, refunds, subscription cancellations.",
        "Can create, update or delete calendar events when you ask ('move my 3pm to tomorrow').",
        "Handles refunds and cancels unused subscriptions on your instruction."
      ],
      "limitations": [
        "Free to start for two weeks, but no paid price published on the pages read 22 October 2026. Unknown pricing is not free.",
        "Broad task claims (refunds, cancellations, insurance checks) were not tested.",
        "Terms note integration with third-party services including Google Calendar and SMS providers; exact scopes were not established."
      ],
      "runtime": [
        "Vendor-managed assistant reached by text message; no app required"
      ],
      "privacy_security": "Del's privacy policy (last updated 30 September 2026) states data from connected integrations is not used to train or improve generalised AI/ML models, and that its AI providers are contractually barred from training general-purpose models on it; Google user data, SMS content, tasks and notes are not used to create, train or improve any AI/ML model beyond personalised features. Google user data is not transferred or sold to third parties including advertisers; Gmail content reaches AI providers only as needed context for a request. Security: encryption in transit and at rest for databases and backups; disconnecting an integration deletes the stored OAuth tokens. Limited human review of requests is permitted and Del discloses in the conversation when a human was involved; the policy says identifiable user content is not reviewed for routine analytics, product development or model training. Its terms include a suicide and self-harm safety protocol (supportive responses, brief safety questions, US 988 crisis-line referrals); Del does not contact emergency services and does not guarantee detection. Checked 24 October 2026.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-24",
        "source_url": "https://withdel.com/",
        "note": "withdel.com (checked 23 October 2026): terms state the service may be free or a paid plan, with fees presented before commitment; no public price list found.",
        "class": "see_vendor"
      },
      "sources": [
        "https://withdel.com/",
        "https://www.withdel.com/terms",
        "https://www.withdel.com/privacy",
        "https://www.withdel.com/use-cases"
      ],
      "changelog": [
        {
          "date": "2026-10-22",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-10-24",
          "change": "Re-verified against current homepage, terms and privacy policy. Added the terms' suicide and self-harm safety protocol and the privacy policy's limited human-review disclosure. Free two-week start confirmed; paid price still unpublished."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Free or paid, amounts unconfirmed",
          "Terms note the service may be offered free or as a paid plan, with fees presented before you commit; no public price list was found on withdel.com as of 23 October 2026.",
          "https://www.withdel.com/terms"
        ],
        [
          "Training use",
          "Not used for general models",
          "Policy: data from connected integrations is not used to train or improve generalised AI/ML models, and providers are contractually barred from training general-purpose models on it. Google user data, SMS content, tasks and notes are not used to train any model beyond personalised features.",
          "https://www.withdel.com/privacy"
        ],
        [
          "Google data",
          "Limited Use",
          "Google user data is not transferred or sold to third parties (including advertising); Gmail content is shared with AI providers only when needed as context for a request.",
          "https://www.withdel.com/privacy"
        ],
        [
          "Security",
          "Encrypted",
          "Encryption in transit and at rest for databases and backups; disconnecting an integration deletes the stored OAuth tokens (Calendar and Gmail).",
          "https://www.withdel.com/privacy"
        ]
      ],
      "setup_steps": [
        "Text Del - no app or login required to start.",
        "Connect Calendar and Gmail for briefings and email work; disconnecting deletes the stored OAuth tokens.",
        "Terms say free or paid plan - confirm any fee prompt in the conversation before committing."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Free 2-week start; paid price not published (site, Oct 2026)",
      "page_url": "https://agentcomparison.net/agents/del/"
    },
    {
      "id": "tinynature",
      "name": "tinyNature",
      "vendor": "Natura AI",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "tinyNature is an assistant you reach on WhatsApp, iMessage or Telegram. It routes requests to internal sub-agents: an executive email agent, a proactive lifestyle manager, background automations (flight updates, briefings, weather), deep web research with cited reports, and real-time tracking of flights, packages and stocks.",
      "one_liner": "A WhatsApp, iMessage and Telegram assistant that delegates requests to internal specialist agents.",
      "best_fit": "People who live in WhatsApp or iMessage and want email triage, reminders, automations and research in one thread.",
      "poor_fit": "Anyone needing published pricing or audited security claims before connecting accounts.",
      "decision_note": "tinyNature sells consolidation: one conversation across three messaging services instead of many apps. It is in beta and publishes no pricing, so cost is unknown. The site claims SOC 2 certification; treat that as a vendor claim until the report is checkable. Sample commands (expense extraction from receipt photos, daily task nudges) suggest broad personal-data access.",
      "availability": "Beta, available now per its site, on WhatsApp, iMessage and Telegram (read 22 October 2026).",
      "last_checked": "2026-10-24",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Email agent: reads, sorts and drafts responses",
          "source_url": "https://tinypeople.ai/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Proactive reminders based on career and habits",
          "source_url": "https://tinypeople.ai/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Background automations: flight updates, daily briefings, weather alerts",
          "source_url": "https://tinypeople.ai/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Deep web research with cited, summarised reports in chat",
          "source_url": "https://tinypeople.ai/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Real-time tracking of flights, packages and stocks",
          "source_url": "https://tinypeople.ai/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Vision tasks: extract expiry dates from grocery photos, receipt expense capture",
          "source_url": "https://tinypeople.ai/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Reads, sorts and drafts email responses via an executive email sub-agent.",
        "Runs proactive reminders, flight/package/stock tracking, daily briefings and custom automations in the background.",
        "Performs deep web research and returns curated, cited summaries in chat.",
        "Accepts smart-vision tasks (e.g. extract expiry dates from grocery photos) and expense triggers (receipts over a threshold saved automatically)."
      ],
      "limitations": [
        "Beta product; no pricing published on the page read 22 October 2026. Unknown pricing is not free.",
        "SOC 2 certification is claimed on the homepage; the report was not inspected.",
        "Sub-agent behaviour, accuracy and reliability were not tested."
      ],
      "runtime": [
        "Vendor-managed assistant on WhatsApp, iMessage and Telegram"
      ],
      "privacy_security": "tinyNature publishes no privacy policy - /privacy and /about 404 as of 23 October 2026. The homepage markets 'Privacy First' with a 'SOC 2 Certified' badge, but links no policy, trust page or certificate, so the claim is unverified vendor marketing. Training use, retention, deletion and sharing practices are unknown. The product's design implies broad access: email reading/drafting, photo uploads and receipt processing all flow through the chat.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": null,
        "usage_charges": null,
        "verified_at": "2026-10-24",
        "source_url": "https://tinypeople.ai/",
        "note": "No pricing published on tinypeople.ai (checked 23 October 2026).",
        "class": "see_vendor"
      },
      "sources": [
        "https://tinypeople.ai/"
      ],
      "changelog": [
        {
          "date": "2026-10-22",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-10-24",
          "change": "Re-verified homepage: still no pricing published; 'Privacy First' SOC 2 badge remains the only privacy artefact (no linked policy, trust page or certificate). No changes to recorded fields."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "Not published",
          "No price, plan or trial information appears on tinypeople.ai as of 23 October 2026.",
          "https://tinypeople.ai/"
        ],
        [
          "Architecture",
          "Sub-agent delegation",
          "One contact delegates to internal sub-agents: executive email agent (reads, sorts, drafts), proactive lifestyle manager (reminders), automations (flight updates, briefings, workflows) and deep web research with cited reports.",
          "https://tinypeople.ai/"
        ],
        [
          "Privacy policy",
          "Not found",
          "tinypeople.ai/privacy and /about 404 as of 23 October 2026. The homepage claims 'SOC 2 Certified' under a 'Privacy First' heading, but no policy or certification detail is linked - treat the claim as unverified vendor marketing.",
          "https://tinypeople.ai/"
        ],
        [
          "Channels",
          "WhatsApp-first",
          "Site title and branding centre WhatsApp; the record also lists iMessage and Telegram entry points. No app install required.",
          "https://tinypeople.ai/"
        ]
      ],
      "setup_steps": [
        "Message tinyNature on WhatsApp (iMessage and Telegram entry points also listed).",
        "No app or settings to configure - describe the task in chat.",
        "Pricing is not published; confirm cost before relying on the service.",
        "No privacy policy is published - weigh that before connecting email or uploading receipts/photos."
      ],
      "price_tags": [
        "unknown"
      ],
      "price_summary": "Beta open now; no pricing published (site, Oct 2026)",
      "page_url": "https://agentcomparison.net/agents/tinynature/"
    },
    {
      "id": "hirealpha",
      "name": "HireAlpha",
      "vendor": "HireAlpha",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "HireAlpha puts three distinct AI personas in your iMessage threads: Friend (sleep, food, training, spending, people you owe replies), Coworker (meetings, drafts, Linear triage, standups) and Cofounder (pipeline, runway, decisions, investor notes). Each has its own phone number, memory and personality, at $19 per persona per month, starting from a waitlist with a Google account connection.",
      "one_liner": "Three AI personas - Friend, Coworker, Cofounder - as separate iMessage contacts at $19/month each.",
      "best_fit": "Founders and operators on iMessage who want separate life-, work- and company-focused assistants with distinct memories.",
      "poor_fit": "Android or WhatsApp users (iMessage only), and anyone wanting one assistant instead of three subscriptions.",
      "decision_note": "The three-persona split is the differentiator: separate numbers and memories instead of one assistant with modes. Pricing is published and simple ($19/persona/month) but access starts from a waitlist, so availability is not immediate. The Google account connection scope and data handling were not established from the pages read.",
      "availability": "Waitlist: start from the waitlist, connect a Google account, then text the personas in iMessage (read 22 October 2026).",
      "last_checked": "2026-10-24",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Friend persona: morning/evening briefs, body tracking, follow-up reminders, weekly review",
          "source_url": "https://hirealpha.chat/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Coworker persona: next-task guidance, meeting prep, draft review, scheduling, issue triage",
          "source_url": "https://hirealpha.chat/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Cofounder persona: pipeline board, decision ledger, hire calls, investor updates",
          "source_url": "https://hirealpha.chat/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Separate phone number, memory and personality per persona",
          "source_url": "https://hirealpha.chat/",
          "observed_at": "2026-10-22",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Friend tracks personal life: sleep, food, training, spending and people you owe a reply.",
        "Coworker runs the workday: next action, meeting prep, drafts and Linear triage.",
        "Cofounder thinks about the company: pipeline, runway, decisions and investor notes.",
        "Each persona has its own number, memory and personality inside your real message threads."
      ],
      "limitations": [
        "Waitlist access; onboarding date not guaranteed (read 22 October 2026).",
        "iMessage only; no Android, WhatsApp or web version documented.",
        "Each persona is a separate $19/month subscription; three personas cost $57/month.",
        "Memory quality and persona reliability were not tested."
      ],
      "runtime": [
        "Vendor-managed personas, each with its own phone number, in iMessage threads"
      ],
      "privacy_security": "HireAlpha's privacy page is short and direct: 'We do not sell your data, advertise against it, or train shared models on it. Your personas read only the Google data you connect and the things you log or text them. Deleting your account deletes your rows within 30 days. Email hello@hirealpha.chat for a copy or deletion.' Retention beyond the 30-day deletion commitment, provider training terms and security detail are not documented further as of 23 October 2026.",
      "pricing": {
        "amount": 19,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Per persona: Friend, Coworker or Cofounder, each $19/month.",
        "usage_charges": null,
        "verified_at": "2026-10-24",
        "source_url": "https://hirealpha.chat/",
        "note": "hirealpha.chat (checked 23 October 2026): $19 per persona per month; three personas (Friend, Coworker, Cofounder).",
        "class": "amount"
      },
      "sources": [
        "https://hirealpha.chat/",
        "https://hirealpha.chat/privacy"
      ],
      "changelog": [
        {
          "date": "2026-10-22",
          "change": "Initial documentation research. No hands-on testing."
        },
        {
          "date": "2026-10-24",
          "change": "Re-verified homepage and privacy page: $19 per persona per month and waitlist-first onboarding unchanged; no-sell/no-ads/no-shared-training and 30-day deletion commitments unchanged."
        }
      ],
      "coverage_rows": [
        [
          "Pricing",
          "$19/persona/mo",
          "$19 per persona per month. Three personas: Friend (sleep, food, training, spending, owed replies), Coworker (next action, meeting prep, drafts, Linear triage), Cofounder (pipeline, runway, decisions, investor notes).",
          "https://hirealpha.chat/"
        ],
        [
          "Channel",
          "iMessage threads",
          "Each persona lives in your real iMessage threads with its own number, memory and personality.",
          "https://hirealpha.chat/"
        ],
        [
          "Training use",
          "None stated",
          "'We do not sell your data, advertise against it, or train shared models on it.'",
          "https://hirealpha.chat/privacy"
        ],
        [
          "Data access",
          "Scoped",
          "Personas read only the Google data you connect and the things you log or text them.",
          "https://hirealpha.chat/privacy"
        ],
        [
          "Deletion",
          "30 days",
          "Deleting your account deletes your rows within 30 days; email hello@hirealpha.chat for a copy or deletion.",
          "https://hirealpha.chat/privacy"
        ]
      ],
      "setup_steps": [
        "Pick a persona - Friend, Coworker or Cofounder - at $19 per persona per month.",
        "Text its number in iMessage; each persona keeps its own memory and personality.",
        "Connect Google data deliberately - personas read only what you connect plus what you log or text.",
        "Deletion: delete the account (rows gone within 30 days) or email hello@hirealpha.chat."
      ],
      "price_tags": [
        "listed"
      ],
      "price_summary": "$19/month per persona; waitlist start (site, Oct 2026)",
      "page_url": "https://agentcomparison.net/agents/hirealpha/"
    },
    {
      "id": "asmi",
      "name": "Asmi",
      "vendor": "Humint Labs, Inc.",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Asmi is an AI assistant from Humint Labs that completes real-world tasks on your behalf: you text it on WhatsApp, SMS or iMessage, and it calls, texts or emails businesses and people - booking restaurants, scheduling appointments, gathering quotes, chasing refunds - then reports back. Persistent memory of your preferences and past tasks is central to the product.",
      "one_liner": "An assistant you text that makes the actual phone calls: bookings, quotes and follow-ups, from $10/month after a free first task.",
      "best_fit": "People who want small real-world errands - bookings, appointments, quotes, follow-ups - handled by an assistant that makes the actual phone calls.",
      "poor_fit": "Anyone wanting a free ongoing tier, guaranteed task completion, or an assistant that never records calls.",
      "decision_note": "First task free, then Pro at $10/month for 20 tasks or Ultra at $49/month for 100 tasks (annual billing is referenced in the terms; prices not captured on this read). Calls on your behalf use an AI voice that identifies itself as an AI and discloses recording; Asmi refuses marketing, debt-collection, emergency, political, harassment and legal-threat calls. It will not make a purchase or binding commitment without the approval the product flow asks for.",
      "availability": "Live per its site; start by texting +1 (408) 668-9990 on WhatsApp or +1 (408) 394-2821 by SMS/iMessage (read 24 October 2026).",
      "last_checked": "2026-10-24",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Calls businesses on your behalf (bookings, appointments, quotes)",
          "source_url": "https://www.asmiai.com/",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Texts and emails third parties, with follow-ups until done",
          "source_url": "https://www.asmiai.com/terms-and-conditions",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Presents options for you to pick before it acts",
          "source_url": "https://www.asmiai.com/",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "AI-voice calls to you for check-ins and briefings",
          "source_url": "https://www.asmiai.com/privacy",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Persistent memory of preferences and past tasks",
          "source_url": "https://www.asmiai.com/privacy",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Refuses restricted call types (marketing, debt collection, emergency, political, harassment, legal threats)",
          "source_url": "https://www.asmiai.com/privacy",
          "observed_at": "2026-10-24",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Contacts third parties only on your explicit per-task instruction; the authorization ends when the task closes (terms).",
        "Calls on your behalf open by identifying Asmi as an AI assistant and disclosing recording; if the recipient objects, recording stops immediately.",
        "Will not knowingly make a purchase or binding commitment without the instruction or approval the product flow requires (terms).",
        "SMS to you is transactional only, sent after consent; STOP and HELP are honoured."
      ],
      "limitations": [
        "Task completion is not guaranteed: third parties can decline, delay or misreport, per the terms.",
        "Only the first task is free; ongoing use is $10/month (20 tasks) or $49/month (100 tasks).",
        "Annual billing is referenced in the terms but annual prices were not captured on this read.",
        "Memory is core to the product: withdrawing consent to call recording means Asmi can no longer function as intended, per the privacy policy.",
        "18+ only. Call quality, task success rate and voice were not tested."
      ],
      "runtime": [
        "Vendor-managed assistant reachable on WhatsApp, SMS and iMessage; it can also place AI-voice calls to you",
        "Account settings on the web for subscription management"
      ],
      "privacy_security": "Privacy policy (effective October 2026): saves conversation records - call audio, transcripts, tasks, preferences and long-term memory. States data is not sold, not used for ad targeting, and not used to train AI models without explicit consent. ElevenLabs processes call audio under a data processing agreement with training disabled; Twilio routes calls without storing content. Retention: recordings 90 days by default, transcripts and memory for the life of the account, deletion within 30 days of a request. CCPA rights stated. Vendor policy, not independently audited.",
      "pricing": {
        "amount": 10,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Pro: 20 tasks per month. Ultra ($49/month): 100 tasks per month. First task free.",
        "usage_charges": null,
        "verified_at": "2026-10-24",
        "source_url": "https://www.asmiai.com/pricing",
        "note": "Pricing page (read 24 October 2026): first task free; Pro $10/month for 20 tasks; Ultra $49/month for 100 tasks. Terms reference annual billing; annual prices not captured. Billed through Stripe.",
        "class": "amount"
      },
      "sources": [
        "https://www.asmiai.com/",
        "https://www.asmiai.com/pricing",
        "https://www.asmiai.com/privacy",
        "https://www.asmiai.com/terms-and-conditions"
      ],
      "changelog": [
        {
          "date": "2026-10-24",
          "change": "Initial documentation research from homepage, pricing, privacy policy and terms of service. No hands-on testing. Completes the 14-agent assistantbenchmark coverage list."
        }
      ],
      "coverage_rows": [
        [
          "Messaging channels",
          "Documented",
          "WhatsApp, SMS and iMessage threads; Asmi can also place AI-voice calls to you.",
          "https://www.asmiai.com/"
        ],
        [
          "Tasks via calls, texts and emails",
          "Documented",
          "Calls, texts and emails third parties on your explicit per-task instruction; presents options before acting.",
          "https://www.asmiai.com/terms-and-conditions"
        ],
        [
          "Purchases and commitments",
          "Documented with approvals",
          "No purchase or binding commitment without the instruction or approval the product flow requires.",
          "https://www.asmiai.com/terms-and-conditions"
        ],
        [
          "Restricted calls",
          "Documented",
          "Refuses marketing, debt-collection, emergency, political, harassment and legal-threat calls.",
          "https://www.asmiai.com/privacy"
        ],
        [
          "Cost",
          "Documented",
          "First task free; Pro $10/month (20 tasks); Ultra $49/month (100 tasks); annual prices not captured.",
          "https://www.asmiai.com/pricing"
        ],
        [
          "Privacy",
          "Documented",
          "Recordings 90 days, memory for account lifetime; no sale, no ad targeting, no training without consent; deletion within 30 days.",
          "https://www.asmiai.com/privacy"
        ]
      ],
      "setup_steps": [
        "Text Asmi on WhatsApp (+1 (408) 668-9990) or SMS/iMessage (+1 (408) 394-2821) and describe the task.",
        "Your first task is free; pick a Pro or Ultra plan for ongoing use.",
        "Give explicit per-task instructions - Asmi contacts third parties only when you ask it to."
      ],
      "price_tags": [
        "listed"
      ],
      "price_summary": "First task free, then $10/month (20 tasks) or $49/month (100 tasks)",
      "page_url": "https://agentcomparison.net/agents/asmi/"
    },
    {
      "id": "stanley",
      "name": "Stanley",
      "vendor": "Find Community, Inc. (dba Stan)",
      "kind": "product",
      "category": "AI content assistant",
      "group": "work-business",
      "summary": "Stanley is Stan's AI head of content for creators. You message it on iMessage, Telegram or the web; it connects your Instagram, LinkedIn and YouTube accounts, turns ideas and voice notes into scripts, captions and posts, plans a content calendar and schedules publishing.",
      "one_liner": "An AI head of content for creators you message on iMessage or Telegram: ideas, scripts and scheduling, $44.99/month after a 3-day trial.",
      "best_fit": "Content creators who want an always-on head of content: ideas, scripts, captions, scheduling and monetisation workflows tied to their Instagram, LinkedIn and YouTube presence.",
      "poor_fit": "Anyone wanting a general personal assistant, a free ongoing tier, or a published web price list.",
      "decision_note": "Stanley is a creator-business tool, not a general assistant: its pitch is consistent posting and monetisation workflows. The iOS app is free with a 3-day trial, then $44.99/month (App Store, October 2026); its own site publishes no prices. Instagram sync reliability is the recurring complaint in App Store reviews - user reports, not our testing.",
      "availability": "Live; message Stanley on iMessage or Telegram, or use the web and iOS app (App Store listing live, read 25 October 2026).",
      "last_checked": "2026-10-25",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Content ideas, hooks, scripts and captions from chat or voice notes",
          "source_url": "https://www.getstanley.ai/",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Content calendar with post scheduling",
          "source_url": "https://www.getstanley.ai/",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Instagram, LinkedIn and YouTube connection for content and engagement context",
          "source_url": "https://www.getstanley.ai/privacy",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Daily nudges, checklists and accountability tracking for posting habits",
          "source_url": "https://apps.apple.com/us/app/stanley-ai-head-of-content/id6754887783",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Monetisation workflows: offers, pricing, funnels and DM scripts",
          "source_url": "https://apps.apple.com/us/app/stanley-ai-head-of-content/id6754887783",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Posts and schedules to connected social accounts on your instruction (vendor site).",
        "YouTube access uses Google API Services: data is displayed only to you, refreshed or deleted per YouTube API policies, deleted when you disconnect, and revocable in Google security settings (privacy policy).",
        "Terms require complying with each connected platform's automation and content rules; spam and platform manipulation are banned."
      ],
      "limitations": [
        "Only a 3-day trial, then $44.99/month; no prices on getstanley.ai itself (terms confirm paid subscriptions; amounts come from the App Store).",
        "App Store reviews repeatedly cite Instagram sync lag and glitches - user reports, not our testing.",
        "Homepage demonstrations (overnight comment replies, brand-deal flagging in DMs) are vendor marketing and were not tested."
      ],
      "runtime": [
        "Messaging threads on iMessage and Telegram, plus web and iOS app.",
        "Connects Instagram, LinkedIn and YouTube accounts for content context, engagement metrics and scheduling."
      ],
      "privacy_security": "Stan platform privacy policy (last revised 27 July 2026; Find Community, Inc. is the EEA/UK controller). Content you submit to Stanley is not used to train general AI models, and third-party AI providers are contractually bound not to train their models on it. Stripe processes payments. The policy allows interest-based advertising: user information may be shared with companies such as Google and Meta to advertise Stan's services. YouTube API data is displayed only to you and deleted on disconnect. Checked 25 October 2026.",
      "pricing": {
        "amount": 44.99,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Stanley Pro monthly subscription; 3-day free trial, then $44.99/month, auto-renewing.",
        "usage_charges": null,
        "verified_at": "2026-10-25",
        "source_url": "https://apps.apple.com/us/app/stanley-ai-head-of-content/id6754887783",
        "note": "App Store listing (read 25 October 2026): free download with in-app purchases; Stanley Pro $44.99/month after a 3-day free trial. getstanley.ai publishes no prices; its terms (updated 22 October 2026) confirm some features require a paid subscription, billed via Stripe, trials converting to paid, cancellable from the Billing page or by messaging Stanley.",
        "class": "amount"
      },
      "sources": [
        "https://www.getstanley.ai/",
        "https://www.getstanley.ai/pricing",
        "https://www.getstanley.ai/privacy",
        "https://www.getstanley.ai/terms",
        "https://apps.apple.com/us/app/stanley-ai-head-of-content/id6754887783"
      ],
      "changelog": [
        {
          "date": "2026-10-25",
          "change": "Initial documentation research from homepage, App Store listing, privacy policy and terms of service. No hands-on testing. Long held (3 October backlog) until vendor pages carried enough detail."
        }
      ],
      "coverage_rows": [
        [
          "Messaging channels",
          "Documented",
          "Two-way threads on iMessage and Telegram, plus web and iOS app.",
          "https://www.getstanley.ai/"
        ],
        [
          "Social account connections",
          "Documented",
          "Instagram content and engagement metrics, LinkedIn content, YouTube channel data via Google API Services.",
          "https://www.getstanley.ai/privacy"
        ],
        [
          "Publishing actions",
          "Documented",
          "Content calendar and post scheduling on your instruction; platform automation rules apply per the terms.",
          "https://www.getstanley.ai/terms"
        ],
        [
          "Cost",
          "Published with limits",
          "$44.99/month after a 3-day trial per the App Store; no prices on the vendor site itself.",
          "https://apps.apple.com/us/app/stanley-ai-head-of-content/id6754887783"
        ],
        [
          "Data use",
          "Policy reviewed",
          "Content not used to train general AI models; providers contractually barred; ads sharing with Google/Meta disclosed.",
          "https://www.getstanley.ai/privacy"
        ]
      ],
      "setup_steps": [
        "Message Stanley on iMessage or Telegram, or install the iOS app.",
        "Connect Instagram (and LinkedIn or YouTube if wanted) for content context.",
        "Start the 3-day trial; cancel from the Billing page, by messaging Stanley, or in App Store settings to avoid the $44.99/month renewal."
      ],
      "price_tags": [
        "listed"
      ],
      "price_summary": "3-day free trial, then $44.99/month (App Store, Oct 2026)",
      "page_url": "https://agentcomparison.net/agents/stanley/"
    },
    {
      "id": "genspark",
      "name": "Genspark",
      "vendor": "Genspark Inc. (MainFunc Inc.)",
      "kind": "product",
      "category": "general-purpose task execution",
      "group": "work-business",
      "summary": "Genspark is an all-in-one AI workspace from MainFunc built around its Super Agent: plain-English requests produce documents, slides, sheets, live dashboards, code, images, video and audio, with specialist tools (AI Docs, Slides and Sheets, Genspark Code, Call For Me, Meeting Notes, SecondBrain) and an AI Drive for storage.",
      "one_liner": "An all-in-one AI workspace whose Super Agent turns plain-English asks into documents, slides and dashboards. Free tier; Plus from $24.99/mo.",
      "best_fit": "People who want one workspace where an agent produces finished artefacts - documents, slides, sheets, dashboards - rather than chat answers.",
      "poor_fit": "Anyone needing guaranteed output accuracy, offline or self-hosted operation, or a messaging-first assistant.",
      "decision_note": "A genuine free tier (100 credits a day) makes it cheap to evaluate; Plus at $24.99/month is the first paid step and Pro at $249.99/month is a steep jump aimed at heavy credit use. Output quality and per-task credit burn were not tested.",
      "availability": "Live on the web; the pricing page requires sign-in, but plan details are published in the vendor's help centre (read 25 October 2026).",
      "last_checked": "2026-10-25",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Super Agent: multi-step tasks producing documents, slides, sheets and live dashboards from plain English",
          "source_url": "https://www.genspark.ai/",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "AI Docs, AI Slides and AI Sheets",
          "source_url": "https://www.genspark.ai/helpcenter/membership-plans",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Genspark Code and GenTerminal for development tasks",
          "source_url": "https://www.genspark.ai/helpcenter/membership-plans",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Call For Me: phone calls placed by the agent",
          "source_url": "https://www.genspark.ai/helpcenter/membership-plans",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Image, video and audio generation; AI Meeting Notes; SecondBrain knowledge base",
          "source_url": "https://www.genspark.ai/",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Chrome extension, Microsoft Office plugin and Salesforce integration",
          "source_url": "https://www.genspark.ai/",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Acts inside the vendor workspace; connectors and integrations such as Salesforce are opt-in per feature.",
        "Google Workspace API data is used only for the requested service under Google's Limited Use policy and is not used to train generalised models (privacy policy).",
        "Queries may be transmitted to OpenAI or Anthropic APIs to serve requests (privacy policy)."
      ],
      "limitations": [
        "Workspace output quality and per-task credit consumption were not tested.",
        "genspark.ai/pricing requires sign-in; plan figures here come from the vendor help centre.",
        "Terms note commercial use of generated content may be limited to paid tiers in future.",
        "Terms ban systematic automated data collection on the service; generated content may not be republished without acknowledging its source."
      ],
      "runtime": [
        "Web workspace at genspark.ai, with a Chrome extension and Microsoft Office plugin.",
        "Tasks run in the vendor environment; AI Drive stores outputs (1 GB free, 50 GB Plus, 1 TB Pro)."
      ],
      "privacy_security": "Privacy policy (updated 24 July 2026): personal data is not sold except in described circumstances; mobile phone numbers and SMS opt-in data are never sold or shared for marketing; queries may be sent to OpenAI or Anthropic APIs; Google Workspace data is handled under the Limited Use requirements with no generalised model training; data is stored on Microsoft Azure. Genspark Community posts remain after account deletion, attributed to an inactive member. Checked 25 October 2026.",
      "pricing": {
        "amount": 24.99,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Free: 100 credits/day, 1 GB AI Drive. Plus: from 10,000 credits/month, 50 GB, commercial use of generated content. Pro: from 125,000 credits/month, 1 TB. Team: per-seat plan.",
        "usage_charges": "Credit-based consumption for AI tasks; a weekly Free Quota covers standard-mode text and image tasks at no credit cost.",
        "verified_at": "2026-10-25",
        "source_url": "https://www.genspark.ai/helpcenter/membership-plans",
        "note": "Vendor help centre, membership plans (read 25 October 2026): Free $0 (100 credits/day, 1 GB); Plus $24.99/month or $19.99/month billed annually ($239.99/year); Pro $249.99/month or $199.99/month annually ($2,399.99/year); Team $40/seat/month or $35/seat/month annually ($420/seat/year); Enterprise on contract. genspark.ai/pricing itself requires sign-in.",
        "class": "free_tier"
      },
      "sources": [
        "https://www.genspark.ai/",
        "https://www.genspark.ai/helpcenter/membership-plans",
        "https://www.genspark.ai/privacy",
        "https://www.genspark.ai/terms"
      ],
      "changelog": [
        {
          "date": "2026-10-25",
          "change": "Initial documentation research from homepage, help-centre membership plans, privacy policy and terms. No hands-on testing. Queued since 3 October backlog."
        }
      ],
      "coverage_rows": [
        [
          "Output artefacts",
          "Documented",
          "Documents, slides, sheets, live dashboards, code, images, video and audio from one workspace.",
          "https://www.genspark.ai/"
        ],
        [
          "Phone calls",
          "Documented",
          "Call For Me places phone calls on your behalf.",
          "https://www.genspark.ai/helpcenter/membership-plans"
        ],
        [
          "Cost",
          "Published",
          "Free $0 (100 credits/day); Plus $24.99/month; Pro $249.99/month; Team $40/seat/month; annual discounts.",
          "https://www.genspark.ai/helpcenter/membership-plans"
        ],
        [
          "Data use",
          "Policy reviewed",
          "No sale of personal data as described; OpenAI/Anthropic process queries; Google Workspace data under Limited Use.",
          "https://www.genspark.ai/privacy"
        ],
        [
          "Access",
          "Documented with limits",
          "Free account on the web; the pricing page itself requires sign-in.",
          "https://www.genspark.ai/helpcenter/membership-plans"
        ]
      ],
      "setup_steps": [
        "Create a free account on genspark.ai (100 credits a day, 1 GB AI Drive).",
        "Run a task in the Super Agent and review the artefacts it produces.",
        "Upgrade to Plus or Pro in the workspace if you need more credits, storage or commercial use."
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Free $0 (100 credits/day); Plus $24.99/mo; Pro $249.99/mo; Team $40/seat/mo",
      "page_url": "https://agentcomparison.net/agents/genspark/"
    },
    {
      "id": "windsurf",
      "name": "Windsurf",
      "vendor": "Cognition AI, Inc.",
      "kind": "product",
      "category": "coding agent (editor and cloud agents)",
      "group": "coding-agents",
      "summary": "Windsurf is Cognition's agentic coding editor: Tab completions, inline edits and agent coding in the IDE, with frontier models (OpenAI, Claude, Gemini), Cognition's own SWE 1.7, and Devin Cloud agents on paid plans.",
      "one_liner": "Cognition's agentic coding editor with frontier models and Devin Cloud agents. Free tier; Pro $20/mo - paid tiers can opt out of training.",
      "best_fit": "Developers who want an agentic IDE with frontier-model choice and are comfortable opting out of training on a paid tier.",
      "poor_fit": "Anyone needing training-free use on a free plan, or a self-hosted editor.",
      "decision_note": "The Free tier is genuinely usable for light agent coding, but note the data trade: Cognition's platform terms say customer data may be used for model training unless you opt out - and only paid tiers can opt out, which then enables zero data retention with model providers. Pro at $20/month is the tier where privacy control starts.",
      "availability": "Live; download the editor or sign in on the web (pricing read 25 October 2026).",
      "last_checked": "2026-10-25",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Agentic coding in the IDE with Tab completions and inline edits",
          "source_url": "https://windsurf.com/pricing",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Frontier model access (OpenAI, Claude, Gemini) plus SWE 1.7 and open-source models",
          "source_url": "https://windsurf.com/pricing",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Full availability from Pro",
          "region_restrictions": null
        },
        {
          "name": "Devin Cloud agents (Pro and above) and Devin Desktop (Team full seats)",
          "source_url": "https://windsurf.com/pricing",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Paid tiers",
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Cloud agents act in the vendor environment; editor runs locally with vendor model calls.",
        "On opt-out (paid tiers): customer data is not used for any other purpose including training, and zero data retention is enabled with model providers; ZDR does not block abuse-flag retention, safety review or legal process (terms 3.3).",
        "Outputs are assigned to the customer (terms 3.1)."
      ],
      "limitations": [
        "Free-tier customer data may be used for model training; the opt-out (with zero data retention) requires a paid tier, and on Teams only an admin can exercise it (platform terms, 30 June 2026).",
        "Editor quality and agent reliability were not tested.",
        "windsurf.com/terms-of-service renders no readable content; the operative terms are Cognition's platform terms, read on cognition.com."
      ],
      "runtime": [
        "Desktop editor (Windsurf) plus Devin Cloud agents on Pro and above; Devin Desktop included for full seats on Team.",
        "Extra usage beyond plan quotas can be purchased at API pricing."
      ],
      "privacy_security": "Cognition privacy policy (updated 9 March 2026) covers Devin and Windsurf: it permits using User Content to train, fine-tune and improve models 'depending on the terms that apply'. The operative split is in the platform terms (updated 30 June 2026): customer data may be used for model training by default; paid-tier customers may opt out, after which data is not used for training and zero data retention applies with model providers (Teams: admin-only opt-out). ZDR exceptions: abuse classifiers, safety/security review, legal process. Outputs are assigned to the customer. Retention otherwise 'as needed', deletion on request. Checked 25 October 2026.",
      "pricing": {
        "amount": 20,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Free: light agent quota, limited models, unlimited inline edits and Tab. Pro: increased quotas, full model availability, SWE 1.7 free, Devin Cloud. Max: significantly higher quotas. Team: $80/month team plan plus $40/month per full dev seat, unlimited flex seats.",
        "usage_charges": "Extra usage purchasable at API pricing on paid plans.",
        "verified_at": "2026-10-25",
        "source_url": "https://windsurf.com/pricing",
        "note": "windsurf.com/pricing (read 25 October 2026): Free $0; Pro $20/month; Max $200/month; Teams $80/month team plan + $40/month per full dev seat; Enterprise custom.",
        "class": "free_tier"
      },
      "sources": [
        "https://windsurf.com/pricing",
        "https://windsurf.com/privacy-policy",
        "https://cognition.com/legal/platform-terms-of-service",
        "https://cognition.com/legal/website-terms-of-use"
      ],
      "changelog": [
        {
          "date": "2026-10-25",
          "change": "Initial documentation research from pricing page, Cognition privacy policy and platform terms of service. No hands-on testing. From the 3 October backlog (coding agents)."
        }
      ],
      "coverage_rows": [
        [
          "Editor and agents",
          "Documented",
          "Agentic IDE with Tab and inline edits; Devin Cloud agents on Pro and above.",
          "https://windsurf.com/pricing"
        ],
        [
          "Cost",
          "Published",
          "Free $0; Pro $20/month; Max $200/month; Team $80/month + $40 per full dev seat; extra usage at API pricing.",
          "https://windsurf.com/pricing"
        ],
        [
          "Training use",
          "Plan-dependent",
          "Customer data may train models by default; paid-tier opt-out stops training and enables ZDR with providers (Teams: admin-only).",
          "https://cognition.com/legal/platform-terms-of-service"
        ],
        [
          "Output ownership",
          "Documented",
          "Outputs assigned to the customer.",
          "https://cognition.com/legal/platform-terms-of-service"
        ]
      ],
      "setup_steps": [
        "Download the Windsurf editor or sign in on the web; the Free tier needs no card.",
        "On any paid tier, exercise the training opt-out (Teams: admin) if you want zero data retention with model providers."
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Free $0; Pro $20/mo; Max $200/mo; Team $80/mo + $40/dev seat",
      "page_url": "https://agentcomparison.net/agents/windsurf/"
    },
    {
      "id": "warp",
      "name": "Warp",
      "vendor": "Denver Technologies, Inc. (d/b/a Warp)",
      "kind": "product",
      "category": "agentic terminal and coding agents",
      "group": "coding-agents",
      "summary": "Warp is an agentic development terminal: a modern terminal with Warp Agent CLI, cloud agents, codebase indexing and a model picker across OpenAI, Anthropic, Gemini, GLM and Kimi models, plus Warp Drive for team workflows.",
      "one_liner": "An agentic development terminal with cloud agents, codebase indexing and model choice. Free tier; Build $20/mo with $20 usage included.",
      "best_fit": "Developers who live in the terminal and want agent runs, codebase indexing and model choice without leaving it.",
      "poor_fit": "Teams needing contractual no-training defaults without configuring data controls, or non-terminal workflows.",
      "decision_note": "The free tier is a real terminal with pay-as-you-go agent usage; Build at $20/month includes $20 of API-rate agent usage, so light users effectively pay for usage they get back. Read the data terms before pointing it at client code: Warp's terms let it train models on licensee data, with data controls configurable per plan.",
      "availability": "Live; download for macOS, Linux or Windows (pricing read 25 October 2026).",
      "last_checked": "2026-10-25",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Modern terminal with Warp Agent CLI and agent runs",
          "source_url": "https://www.warp.dev/pricing",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Cloud agents and any-harness cloud runs (beta)",
          "source_url": "https://www.warp.dev/pricing",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Extended access from Build",
          "region_restrictions": null
        },
        {
          "name": "Model picker across OpenAI, Anthropic, Gemini, GLM and Kimi models",
          "source_url": "https://www.warp.dev/pricing",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Codebase indexing and Warp Drive team collaboration",
          "source_url": "https://www.warp.dev/pricing",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Highest indexing limits from Build",
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Cloud agents act in the vendor environment; bring-your-own inference endpoints are supported on paid plans (pricing page).",
        "Data controls: individually configured on Free/Build, admin-configured on Business (pricing page).",
        "Terms ban scraping and using the services or documentation to build a competing product."
      ],
      "limitations": [
        "Terms grant Warp the right to use licensee data, including for training its algorithms and models, with third-party disclosure only in aggregate or de-identified form; data controls are configurable per plan (individually on Free/Build, admin on Business) - verify your setting before sensitive code.",
        "Agent output quality and usage burn were not tested.",
        "Subscriptions are non-refundable except for material breach (terms)."
      ],
      "runtime": [
        "Desktop terminal app (macOS, Linux, Windows) with Warp Agent CLI; cloud agents run in the vendor environment.",
        "Add-on usage is shared across a team and rolls over month to month (pricing page)."
      ],
      "privacy_security": "Privacy notice (last modified 18 August 2026): personal information may be used to develop and improve products, including training, fine-tuning and improving the performance of models that support its AI features; marketing and interest-based advertising disclosed. The terms (read 25 October 2026) grant Warp a licence to use licensee data for analytics and product improvement, including training its algorithms and models, with third-party disclosure only in aggregate or de-identified form. Data controls are configurable (individually on Free/Build, admin on Business). Retention is criteria-based; deletion per the notice. Checked 25 October 2026.",
      "pricing": {
        "amount": 20,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Free: terminal + Agent CLI, pay-as-you-go usage (free-plan rates 20% above paid), BYO inference, limited cloud agents. Build: $20 of agent usage/month at API rates. Max: $200 of usage/month. Business: $20 of usage per seat/month (up to 25 seats).",
        "usage_charges": "Agent usage metered at API rates; included amounts per plan; add-on usage rolls over month to month and is shared across a team.",
        "verified_at": "2026-10-25",
        "source_url": "https://www.warp.dev/pricing",
        "note": "warp.dev/pricing (read 25 October 2026): Free $0; Build $20/month or $18/month billed annually; Max $200/month or $180/month annually; Business $50/user/month or $45/user/month annually; Enterprise custom.",
        "class": "free_tier"
      },
      "sources": [
        "https://www.warp.dev/pricing",
        "https://www.warp.dev/privacy",
        "https://www.warp.dev/terms-of-service"
      ],
      "changelog": [
        {
          "date": "2026-10-25",
          "change": "Initial documentation research from pricing page, privacy notice and software/services licence terms. No hands-on testing. From the 3 October backlog (coding agents)."
        }
      ],
      "coverage_rows": [
        [
          "Terminal and agents",
          "Documented",
          "Modern terminal, Agent CLI, cloud agents and any-harness runs (beta).",
          "https://www.warp.dev/pricing"
        ],
        [
          "Cost",
          "Published",
          "Free $0; Build $20/month ($20 usage included); Max $200/month ($200 usage); Business $50/user/month; usage at API rates, rolls over.",
          "https://www.warp.dev/pricing"
        ],
        [
          "Training use",
          "Configurable",
          "Terms let Warp train on licensee data; disclosure to third parties only aggregate/de-identified; data controls per plan.",
          "https://www.warp.dev/terms-of-service"
        ],
        [
          "Refund rule",
          "Documented",
          "Subscriptions non-refundable except for Warp's material breach.",
          "https://www.warp.dev/terms-of-service"
        ]
      ],
      "setup_steps": [
        "Download Warp for macOS, Linux or Windows; the free tier includes the terminal and Agent CLI with pay-as-you-go usage.",
        "Check your data-control setting (individual on Free/Build) before pointing agents at sensitive code."
      ],
      "price_tags": [
        "free",
        "listed"
      ],
      "price_summary": "Free $0; Build $20/mo; Max $200/mo; Business $50/user/mo",
      "page_url": "https://agentcomparison.net/agents/warp/"
    },
    {
      "id": "skye",
      "name": "Skye",
      "vendor": "Signull Labs, Inc.",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Skye is an ambient personal AI agent from Signull Labs for iPhone: it connects calendar, Gmail, spending via Plaid, Apple Health and Whoop, Slack, location and contacts into one ranked feed of cards on the home and lock screen, with a morning audio briefing and a chat that answers from the connected data.",
      "one_liner": "An ambient iPhone agent that ranks your calendar, mail, spending and health into one home-screen feed. Free.",
      "best_fit": "iPhone users who want an ambient, ranked feed and morning briefing across calendar, mail, spending and health, rather than a chat-first assistant.",
      "poor_fit": "Anyone not on iOS 26 or later, Android users, or people wanting a messaging-first assistant or guaranteed inference accuracy.",
      "decision_note": "Free on the App Store with no in-app purchases listed (25 October 2026); the terms say paid plans or per-feature pricing may follow with at least 30 days notice. Agentic actions across connected services are confirm-first per the terms. Feed ranking quality and inference accuracy were not tested.",
      "availability": "Live on the Apple App Store (free, version 1.0.8, requires iOS 26 or later) after an invitation-only beta. Read 25 October 2026.",
      "last_checked": "2026-10-25",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Ambient ranked feed of contextual cards on the iPhone home and lock screen, built from connected calendar, email, spending, health and location data",
          "source_url": "https://skyeapp.ai/",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Morning briefing (audio and feed) with schedule, sleep, news and one forgotten item",
          "source_url": "https://apps.apple.com/us/app/skye-personal-ai-agent/id6748042747",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Chat assistant that answers questions across connected data sources",
          "source_url": "https://apps.apple.com/us/app/skye-personal-ai-agent/id6748042747",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Confirm-first agentic actions: draft or send messages, schedule or change meetings, set reminders and alarms (terms 1.4)",
          "source_url": "https://skyeapp.ai/terms",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Spending context: transactions paired with receipts from the inbox; subscription renewals and fees surfaced",
          "source_url": "https://apps.apple.com/us/app/skye-personal-ai-agent/id6748042747",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Daily Energy Score from Apple Health and Whoop sleep, strain, steps and workouts",
          "source_url": "https://apps.apple.com/us/app/skye-personal-ai-agent/id6748042747",
          "observed_at": "2026-10-25",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Agentic actions across connected services - drafting or sending messages, scheduling or changing meetings, setting reminders - are presented for confirmation before execution (terms 1.4, read 25 October 2026); a future opt-in mode for routine actions without per-action confirmation stays off until enabled.",
        "Plaid connections are read-only: the terms state no payment, transfer or other write-capable permission is requested and money cannot be moved.",
        "Each data connection is granted separately and can be disconnected at any time (App Store listing and privacy policy)."
      ],
      "limitations": [
        "iOS 26 or later only; no Android or web version documented.",
        "Inferences are stored as Memories and Relationships with quoted source excerpts and may be wrong (privacy policy).",
        "Mixpanel analytics is pseudonymous, not anonymous: event labels can include a nearby place name or a suggested prompt drawn from user data.",
        "Financial sync runs through Plaid with independent retention under Plaid’s own policy.",
        "Feed ranking quality, briefing quality and inference accuracy were not tested."
      ],
      "runtime": [
        "iPhone and iPad app (iOS/iPadOS 26 or later), with home-screen and lock-screen widgets and daily audio briefings.",
        "Connections: Google Calendar, Gmail, Slack, Apple Health, Whoop, Plaid (read-only financial), location and contacts; each optional and revocable."
      ],
      "privacy_security": "Privacy policy (dated 09/08/2026, read 25 October 2026): data is not sold and not used for advertising. AI processing runs through OpenAI (primary), Anthropic, Google Gemini (fallback), Groq and ElevenLabs, each listed with zero data retention and no model training. Most connected-service data is retained for no more than 90 days; live-mirror data persists while its source stays connected; inferred Memories and Relationships persist until the user asks for deletion. Crash reporting (Sentry) blacks out all text and images on captured screens before they leave the device, unless the user attaches a screenshot to feedback. Mixpanel product analytics is pseudonymous rather than anonymous.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": "App Store listing: Free; no in-app purchases shown (read 25 October 2026).",
        "usage_charges": null,
        "verified_at": "2026-10-25",
        "source_url": "https://apps.apple.com/us/app/skye-personal-ai-agent/id6748042747",
        "note": "Free on the App Store with no in-app purchases listed (25 October 2026). Terms 6.1: the app currently offers certain features free of charge and may introduce paid plans or per-feature pricing with at least 30 days notice.",
        "class": "free"
      },
      "sources": [
        "https://skyeapp.ai/",
        "https://apps.apple.com/us/app/skye-personal-ai-agent/id6748042747",
        "https://skyeapp.ai/privacy",
        "https://skyeapp.ai/terms"
      ],
      "changelog": [
        {
          "date": "2026-10-25",
          "change": "Initial documentation research from homepage, App Store listing, privacy policy and terms. Held since early October as invitation-only beta; public App Store launch confirmed 25 October 2026. No hands-on testing."
        }
      ],
      "price_tags": [
        "free"
      ],
      "page_url": "https://agentcomparison.net/agents/skye/"
    },
    {
      "id": "claude-duet",
      "name": "Claude Duet",
      "vendor": "Anthropic",
      "kind": "product",
      "category": "proactive work agent",
      "group": "work-business",
      "summary": "Claude Duet is Anthropic’s proactive agent: you hand off work by text or voice message, it keeps working on its own cloud computer while you are away, and it comes back with finished docs, sheets, decks or dashboards - or asks when it needs your approval. It shares your Claude account’s memory, skills and 6,000+ connectors.",
      "one_liner": "Anthropic’s proactive agent: morning briefs and background tasks on its own cloud computer. Pro/Max, US research preview.",
      "best_fit": "Professionals and small-business owners already paying for Claude who want an agent that briefs them, chases work in the background and drafts for approval.",
      "poor_fit": "Anyone outside the US, on a free or Team/Enterprise plan, or needing errands, bookings or live browser transparency - Anthropic says Duet is not built for those.",
      "decision_note": "No extra charge beyond a Claude Pro ($20/month) or Max (from $100/month) subscription, but it is a US-only research preview reached through a waitlist, and its background work counts against the same plan usage limits. Approval-first design with hard limits on money movement. Output quality untested.",
      "availability": "Research preview for Claude Pro and Max subscribers in the US, via waitlist, on iOS, Android and the web. Team and Enterprise plans not included yet. Read 26 October 2026.",
      "last_checked": "2026-10-26",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Proactive morning brief and attention-flagging across connected inbox and calendar",
          "source_url": "https://claude.com/product/claude-duet",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Pro or Max; US research preview",
          "region_restrictions": "US only"
        },
        {
          "name": "Background task execution on its own cloud computer, returning docs, sheets, decks and dashboards",
          "source_url": "https://claude.com/product/claude-duet",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Small-business operations: finding unpaid invoices and drafting reminders, payroll preparation for review, contract preparation",
          "source_url": "https://support.claude.com/en/articles/17631495-get-started-with-claude-duet",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "6,000+ connectors and plugins via the Claude Marketplace",
          "source_url": "https://claude.com/product/claude-duet",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Purchases with one-time merchant-locked wallet cards, approved by the user in the wallet app",
          "source_url": "https://claude.com/resources/articles/claude-duet-safety-privacy-and-security",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Approval-first for sensitive actions: before sending a message, paying or deleting, it shows the action, destination and content including every recipient, and waits (vendor safety page, read 26 October 2026).",
        "Some actions can never skip approval (paying or sending money, deleting an order, creating an account); securities trading, inter-account transfers and accessing others’ accounts are off-limits entirely.",
        "Purchases use one-time merchant-and-amount-locked cards from a wallet such as Link from Stripe, approved in the wallet app; the real card number is never shared with Duet or the merchant.",
        "Passwords, cookies and sign-in tokens never reach the model; they pass through separate machines the model cannot access."
      ],
      "limitations": [
        "US-only research preview behind a waitlist; availability and behaviour may change.",
        "Background checks and larger tasks consume the same Claude plan usage limits as conversations.",
        "Cannot trade securities, move money between accounts or access other people’s accounts, even with approval.",
        "Anthropic acknowledges its outbound-message safety classifier will occasionally miss something or over-flag during the preview.",
        "Output quality, task reliability and approval cadence were not tested."
      ],
      "runtime": [
        "iOS and Android apps plus the web via claude.ai; works on its own cloud computer in the background.",
        "Uses the Claude account’s memory, skills and connectors (6,000+ via the Claude Marketplace)."
      ],
      "privacy_security": "Vendor safety page (read 26 October 2026): approvals and safety checks run outside the model on separate machines; an outbound-message classifier screens drafts, with “never share” treatment for credentials, card and ID numbers, and consent-based sharing for health or finance details. Conversation history and memory persist until the user deletes them; deleting the account wipes everything Duet stored. Anthropic cites external penetration testing and red-teaming of credential features. No statement on model training was found on the Duet safety page; the general Claude privacy terms apply.",
      "pricing": {
        "amount": 20,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Included with Claude Pro or Max during the research preview; background checks count against plan usage limits.",
        "usage_charges": "Shares the host plan’s usage limits; larger tasks consume more.",
        "verified_at": "2026-10-26",
        "source_url": "https://claude.com/pricing",
        "note": "Requires Claude Pro ($20/month, $17/month billed annually) or Max (from $100/month) - read 26 October 2026. Duet itself has no separate price; research preview, waitlist, US only.",
        "class": "amount"
      },
      "sources": [
        "https://claude.com/product/claude-duet",
        "https://claude.com/resources/articles/claude-duet-safety-privacy-and-security",
        "https://support.claude.com/en/articles/17631495-get-started-with-claude-duet",
        "https://claude.com/pricing"
      ],
      "changelog": [
        {
          "date": "2026-10-26",
          "change": "Initial documentation research from the product page, safety/privacy/security page, help centre and pricing page. Launched 22 October 2026 as a US research preview. No hands-on testing."
        }
      ],
      "price_tags": [
        "listed"
      ],
      "page_url": "https://agentcomparison.net/agents/claude-duet/"
    },
    {
      "id": "sundae",
      "name": "Sundae",
      "vendor": "Sundae Bar PLC (AIM: SBAR)",
      "kind": "product",
      "category": "personal assistant",
      "group": "personal-assistants",
      "summary": "Sundae is sundae_bar’s personal agent for everyday life: you text it on iMessage or RCS and it answers in the same thread. It books tables, compares prices, sets reminders, monitors prices on a schedule and takes bigger jobs like flights or broadband deals into the background, texting you once when done.",
      "one_liner": "A text-message personal agent from sundae_bar: text it tasks over iMessage or RCS. Free daily limit in beta; Plus plan coming.",
      "best_fit": "People who want a text-first assistant for everyday jobs - bookings, price checks, reminders - with no app to install and no card to start.",
      "poor_fit": "Anyone needing guaranteed beta access, a published paid price before relying on it, or an assistant that completes purchases for them - Sundae deliberately stops at the checkout.",
      "decision_note": "Free to start with a daily limit and no app, account or card needed - a genuinely low-friction try. It is in beta with limited places, and the paid Plus plan’s price is unpublished. Purchases are confirm-first: Sundae stops at the checkout and you type the card yourself.",
      "availability": "Open beta with limited places: text +1 (937) 896-2734 on iMessage or RCS; if there is no reply, the beta is full. Read 26 October 2026.",
      "last_checked": "2026-10-26",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Text-message tasks over iMessage or RCS: bookings, price checks, reminders, quick questions",
          "source_url": "https://sundae.md/",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Background jobs for bigger tasks such as flights and broadband deals, with a single text when done",
          "source_url": "https://sundae.md/",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Scheduled price monitoring and shopping comparisons, larger results returned as a spreadsheet",
          "source_url": "https://sundae.md/",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Restaurant and booking comparisons with a recommendation, booking only on user confirmation",
          "source_url": "https://sundae.md/",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Persistent memory of what the user tells it, viewable and editable by asking the agent",
          "source_url": "https://www.sundaebar.ai/privacy-policy",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Confirm-first money and external actions: before spending, sending anything to someone else, deleting anything or opening an account, it asks; it stops at the checkout and the user types the card and presses pay (sundae.md, read 26 October 2026).",
        "It will not type passwords or card details for the user.",
        "Bookings are made only when the user says so."
      ],
      "limitations": [
        "Beta with limited places; access can silently be full.",
        "Plus plan announced but its price is unpublished.",
        "The vendor’s team can see conversations to keep the service running (site notice).",
        "Model providers may change automatically mid-conversation; providers are named on a sub-processors page.",
        "Booking accuracy and task completion were not tested."
      ],
      "runtime": [
        "iMessage or RCS to +1 (937) 896-2734 - no app, account or card required to start.",
        "Runs on sundae_bar’s own agent platform with its own computer, browser and memory."
      ],
      "privacy_security": "Privacy policy (read 26 October 2026): model providers are not permitted to train on conversations; providers may retain data briefly for security and abuse monitoring and can change automatically (named on a sub-processors page). The vendor’s team can see conversations to operate the service; administrator access is restricted. Agent memory can be viewed and changed by asking the agent and is deleted on account deletion; Explore-page conversations delete after 30 days of inactivity; payment and tax records are kept seven years. Marketing cookies and third-party marketing shares require consent.",
      "pricing": {
        "amount": null,
        "currency": null,
        "billing_period": null,
        "allowances": "Free to start with a daily limit that resets at midnight local time; Sundae warns when close to the limit.",
        "usage_charges": "A paid plan called Plus raises the daily limit; price unpublished.",
        "verified_at": "2026-10-26",
        "source_url": "https://sundae.md/",
        "note": "Free to start with a daily limit (read 26 October 2026); Plus plan announced but price not yet published. Beta, limited places.",
        "class": "free_start_claimed"
      },
      "sources": [
        "https://sundae.md/",
        "https://www.sundaebar.ai/agent/sundae",
        "https://www.sundaebar.ai/privacy-policy",
        "https://www.sundaebar.ai/news/a-letter-from-the-ceo-meet-sundae"
      ],
      "changelog": [
        {
          "date": "2026-10-26",
          "change": "Initial documentation research from sundae.md, the sundae_bar agent page, privacy policy and CEO launch letter. Launched via RNS 21 October 2026; beta with limited places. No hands-on testing."
        }
      ],
      "price_tags": [
        "free"
      ],
      "page_url": "https://agentcomparison.net/agents/sundae/"
    },
    {
      "id": "eney",
      "name": "Eney",
      "vendor": "MacPaw",
      "kind": "product",
      "category": "proactive desktop assistant",
      "group": "personal-assistants",
      "summary": "Eney is MacPaw’s proactive AI assistant that lives on Mac and iPhone: it understands the apps, files and calendar on the machine itself, spots what needs attention, prepares the next step and asks before it acts. Give it a task in plain words - by typing or voice - and it works across Apple Mail, Calendar, Reminders and connectors for Gmail, Google Drive, Slack, Notion, Jira and Linear.",
      "one_liner": "MacPaw’s proactive assistant for Mac and iPhone: spots what needs attention across your apps and files, then asks before it acts.",
      "best_fit": "Mac and iPhone users - freelancers, assistants, managers - who want a proactive assistant that works across their local apps and files with confirm-first actions.",
      "poor_fit": "Windows or Android users, Intel Macs, anyone wanting a free-forever tier (trial only), or an assistant that acts autonomously without per-action approval.",
      "decision_note": "A Mac-native assistant from an established Mac vendor (since 2008), with an unusually clear permission model: proactive, never autonomous by default. Priced on AI credits - $14.99/month for 10,000 after a 14-day free trial. Mac-only (Apple Silicon, macOS 15+) and English-plus-7-languages; output quality untested.",
      "availability": "Available now: Mac app from eney.ai (also via Setapp) and iPhone on the App Store. Requires Apple Silicon Mac with macOS 15 or later. Launched 15 October 2026. Read 26 October 2026.",
      "last_checked": "2026-10-26",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Proactive suggestions from on-device context: surfaces what needs attention across apps, files and calendar and prepares the next step",
          "source_url": "https://macpaw.com/news/eney-announcement",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Meeting transcription without a bot in the call, with drafted notes, action items and follow-up invite for approval",
          "source_url": "https://macpaw.com/news/eney-announcement",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Eney Wand: Option-W drag over anything on screen to translate, convert, copy text from images, create calendar events or blur sensitive content",
          "source_url": "https://macpaw.com/news/eney-announcement",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Ready-made workflows (Morning Brief, Plan My Day, Weekly Review) run on demand or on a schedule",
          "source_url": "https://macpaw.com/eney",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Mac care: low-disk and memory-pressure warnings, tidy Downloads, fast file search, file conversion, PDF splitting, image/video edits",
          "source_url": "https://macpaw.com/eney",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Cross-app work through connectors: draft emails, pull unread messages, update work in Gmail, Drive, Slack, Notion, Jira and Linear",
          "source_url": "https://macpaw.com/eney",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Confirm-first by default: Eney asks before it sends a message, changes an event or moves a file, and the user decides how much it can do on its own (launch announcement, 15 October 2026).",
        "File and app access only with user permission (product FAQ); the Mnemos memory layer manages only Eney’s own memory and never reads or changes user files or other apps (vendor claim).",
        "Access to connected Google services only with consent; vendor states Google data is not used for advertising, profiling, selling or training external AI models."
      ],
      "limitations": [
        "Mac and iPhone only: Apple Silicon, macOS 15 or later.",
        "No free tier after the 14-day trial: continued use needs a Setapp AI-credits subscription from $14.99/month.",
        "Usage is metered in AI credits; bigger tasks consume more, so heavy use may need the $29.99 or $49.99 tier.",
        "The privacy notice says anonymized usage data may train MacPaw’s local ML models (vendor states identifiers are removed first).",
        "Output quality, transcription accuracy and connector reliability were not tested."
      ],
      "runtime": [
        "Native Mac app (Apple Silicon, macOS 15+) plus iPhone app; opened from anywhere with a double-tap of Option; text or voice input in eight languages.",
        "Connectors: Apple Mail, Calendar and Reminders built in; Gmail, Google Calendar, Google Drive, Slack, Notion, Jira, Linear and more."
      ],
      "privacy_security": "Eney Privacy Notice (updated 24 June 2026, read 26 October 2026): requests route through MacPaw’s backend to Anthropic, Google or OpenAI as data processors, which by default do not train on user data. MacPaw itself may use anonymized User Input, Interaction, Output and Task Execution data to train its own local ML models after removing direct identifiers. Google-integration data is not used for ads, profiling, sale or external AI training. Users can request data deletion; the vendor warns that feedback submissions are human-reviewed.",
      "pricing": {
        "amount": 14.99,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Starter: 10,000 AI credits/month. Boost: 25,000 credits at $29.99/month. Power: 50,000 credits at $49.99/month. Prices include tax, billed monthly.",
        "usage_charges": "AI-credit metered: simple tasks use few credits, bigger ones more.",
        "verified_at": "2026-10-26",
        "source_url": "https://macpaw.com/eney",
        "note": "Free 14-day trial with 10,000 AI credits, then a Setapp AI-credits subscription from $14.99/month (read 26 October 2026). Setapp sign-in required after trial; Setapp membership not required.",
        "class": "amount"
      },
      "sources": [
        "https://macpaw.com/news/eney-announcement",
        "https://macpaw.com/eney",
        "https://macpaw.com/eney-privacy-policy"
      ],
      "changelog": [
        {
          "date": "2026-10-26",
          "change": "Initial documentation research from the launch announcement, product page and Eney Privacy Notice. Launched 15 October 2026. No hands-on testing."
        }
      ],
      "price_tags": [
        "listed"
      ],
      "page_url": "https://agentcomparison.net/agents/eney/"
    },
    {
      "id": "gemini-enterprise",
      "name": "Gemini Enterprise",
      "vendor": "Google Cloud",
      "kind": "product",
      "category": "enterprise work agent",
      "group": "work-business",
      "summary": "Gemini Enterprise is Google’s universal agent for work: a single prompt box that answers questions, handles knowledge work, creates images and media, and writes and runs code. You delegate an outcome and it plans the work, uses skills and tools, connects to your systems and returns finished work inside the documents, inbox and developer environments you already use.",
      "one_liner": "Google’s universal agent for work: one prompt box for knowledge work, media and code, with enterprise governance. From $21/seat/month.",
      "best_fit": "Organizations that want one governed agent across knowledge work, content and coding, with model choice, cost controls and enterprise administration, especially those already on Google Workspace or Google Cloud.",
      "poor_fit": "Individuals wanting a consumer assistant (Google’s consumer Gemini is a different product), and teams needing transparent self-serve pricing above the Business edition.",
      "decision_note": "Google’s flagship work agent, announced at Gemini at Work on 8 October 2026, with the governance enterprises ask for: no training on customer data without permission, Model Armor protections and cost controls. Self-serve Business edition starts at $21 per seat per month with a 30-day trial; larger editions are sales-led. Real-world agent quality untested.",
      "availability": "Business edition self-serve for teams up to 300 with a 30-day trial; Standard and Plus editions via sales. Access from web, iOS, Android, Windows, Mac, command line, Google Workspace, Microsoft 365, Slack and headless API. Industry specializations in preview for Financial Services and Legal. Read 26 October 2026.",
      "last_checked": "2026-10-26",
      "evidence_status": "documentation_researched_not_tested",
      "capabilities": [
        {
          "name": "Single universal agent for knowledge work, Q&A, content creation and coding from one prompt box, with delegated outcomes",
          "source_url": "https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Business-context connection through tools, skills and memory across systems of record",
          "source_url": "https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Connectors for Microsoft 365, Google Workspace, Jira, GitHub, HubSpot, Slack and more",
          "source_url": "https://cloud.google.com/gemini-enterprise/business",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Business edition includes select connectors",
          "region_restrictions": null
        },
        {
          "name": "Per-job model orchestration across Gemini and Anthropic Claude models for cost and quality",
          "source_url": "https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": null,
          "region_restrictions": null
        },
        {
          "name": "Industry specializations for Financial Services and Legal (preview)",
          "source_url": "https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026",
          "observed_at": "2026-10-26",
          "evidence_type": "vendor_documentation",
          "confidence": "medium",
          "plan_restrictions": "Preview",
          "region_restrictions": null
        }
      ],
      "permissions": [
        "Agent works under enterprise identity and policy management with authorization and permission controls, secure sandboxing and network gateways (vendor blog, 8 October 2026).",
        "Can act as a personal assistant, a team member working for a group, or a role-based agent; tasks can be assigned, scheduled or event-triggered.",
        "Google will not use customer data to train or fine-tune AI/ML models without prior permission (Service Specific Terms, per vendor documentation)."
      ],
      "limitations": [
        "Standard and Plus edition pricing is not published - contact sales.",
        "Grounding with Google Search stores derived query logs for up to 3 days and cannot be disabled without switching to Web Grounding for Enterprise (vendor documentation).",
        "Advanced AI features carry abuse-monitoring prompt and response logging, so zero data retention is not always possible (vendor documentation).",
        "Industry specializations beyond Financial Services and Legal are announced as coming soon, not available.",
        "Agent output quality and task completion were not tested."
      ],
      "runtime": [
        "Web, iOS and Android apps, Windows and Mac desktops, command line, inside Google Workspace, Microsoft 365 and Slack, plus a headless API for embedding.",
        "Model choice per job: orchestrates across the Gemini model family and Anthropic Claude models today, with other models planned (vendor blog)."
      ],
      "privacy_security": "Vendor documentation (read 26 October 2026): a Training Restriction in the Service Specific Terms bars Google from using customer data to train or fine-tune AI/ML models without prior permission, covering GA and pre-GA models on the Gemini Enterprise Agent Platform. Abuse-monitoring logging applies to some Advanced AI features; Grounding with Google Search stores de-associated query logs up to 3 days. Business edition includes Model Armor protections. Enterprise identity, policy and authorization controls govern what agents can do.",
      "pricing": {
        "amount": 21,
        "currency": "USD",
        "billing_period": "month",
        "allowances": "Business edition from $21 per seat per month for teams up to 300; includes the Gemini Enterprise app, Notebook and select business connectors (Microsoft 365, Google Workspace, Jira, GitHub, HubSpot, Slack).",
        "usage_charges": "Flexible spending options and built-in cost controls announced; per-seat licensing with consumption-based options per vendor.",
        "verified_at": "2026-10-26",
        "source_url": "https://cloud.google.com/gemini-enterprise/business",
        "note": "From $21/seat/month Business edition with a 30-day free trial (read 26 October 2026); Standard and Plus editions are contact-sales.",
        "class": "amount"
      },
      "sources": [
        "https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026",
        "https://cloud.google.com/gemini-enterprise/business",
        "https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/zero-data-retention"
      ],
      "changelog": [
        {
          "date": "2026-10-26",
          "change": "Initial documentation research from the Gemini at Work 2026 announcement, Business edition page and zero-data-retention documentation. Announced 8 October 2026. No hands-on testing."
        }
      ],
      "price_tags": [
        "listed"
      ],
      "page_url": "https://agentcomparison.net/agents/gemini-enterprise/"
    }
  ]
}