AgentComparison
Personal agents ยท Documentation-led guide

Instinct vs Muse vs OpenAI dots

Instinct describes a text-and-call personal assistant; OpenAI dots connects to ChatGPT plugins and memory; Meta's Muse works through an app or web chat with reminders and a file library. For sensitive connected-account work, the clearest documented differences are how permissions, training and deletion work, not a tested winner on reliability.

Dots and Muse sources rechecked 5 November 2026; Instinct policy review 5 November in linked comparisonNo hands-on testVendor statements, not audits

Instinct vs Dots: the short answer

Consider Instinct if you want to reach an assistant by text or call. Consider OpenAI dots if sharing context with ChatGPT and using its permitted plugins matters more. OpenAI publishes more specific approval and memory-deletion guidance in the pages reviewed here. That is a difference in public documentation, not proof that dots is safer or follows through better.

Neither should be chosen on an assumed free price. Instinct's terms describe possible fees in USD without a universal price. OpenAI documents Pro and Business Premium access separately from Enterprise billing. Pro rollout excludes the EEA, Switzerland and the UK; Business Premium eligibility is region/workspace dependent. Eligible Enterprise workspaces have a free preview through 18 November, when billing begins for enabled dots. Work delegated to Work or Codex is billed separately. Confirm your actual offered access and cost before connecting accounts.

Compare the documented differences

DecisionInstinctOpenAI dotsMuse
How you reach itText or call, according to its homepage.ChatGPT-linked assistant; the desktop app connects messaging channels such as Slack or texting (docs updated 28 October 2026).App or web chat with a Meta Account.
Connected accessConnected-service data and actions under its terms; review each account's settings.Plugin permissions shared across dots, ChatGPT, ChatGPT Work and Codex; device and cloud access have separate controls.Many connectors support retrieve-only access. Facebook, Instagram and Threads connect automatically when accounts share an Accounts Center. Custom connectors are not reviewed by Meta; inspect their provider and permissions (help re-read 8 November 2026).
ApprovalTerms allow responsive actions and warn safeguards are not guaranteed. A complete per-action approval table is not stated there.Custom Rules set boundaries but cannot override core safeguards. Some sensitive actions require takeover; others may need approval.Connector and website defaults are separate. Connector Ask for some actions covers every write and important read; Always ask covers any connector action. Approval scopes include once, task, site, always or deny (help re-read 8 November 2026).
Conversation memoryPrivacy policy describes personalisation using interactions and connected information. Per-memory editing was not established in the pages reviewed.Shares memory with ChatGPT. Turning off ChatGPT Memory stops sharing, not deletion of already received context. Individual dot memories cannot currently be viewed or edited.Remembers conversation context across chats; Meta documents editable Memory.md, Soul.md and Identity.md files plus a forget skill (read 28 October 2026).
Disconnect and deleteDisconnecting does not delete indexed connected-service input. Terms point to Workspace for deletion requests.Disconnecting stops new access, not retained context. Deleting a dot starts deletion of its own context; separately manage ChatGPT Memory, files, conversations and scheduled work.Meta documents deletion of messages, side chats and files, plus reset of all Muse data. Removing a source may leave learned memories; the forget skill removes related memories/files to the best of its ability, not as a guaranteed erasure audit.
Model trainingTerms permit training with a settings opt-out and safety-review exception; Vault materials are excluded. Privacy policy separately excludes information received directly through Google Workspace APIs from model training and improvement.Business, Enterprise and Edu data is not used for training by default. Personal plans use the "Improve the model for everyone" setting. Background research notes are not directly trained on, but information brought into eligible conversations can be.Meta documents a training toggle that is on by default with a retroactive opt-out, and says it does not share data with advertisers (privacy help read 28 October 2026).
Published cost in these sourcesNo universal price; possible USD fees.Pro and Business Premium access is plan/region dependent. Eligible Enterprise preview is free 4-18 November 2026; billing starts 18 November for enabled dots. Flat-rate or scaled consumption; delegated Work/Codex billed separately. Exact dollar rates not established here.Free with a weekly usage limit (size unpublished); Power $20/month (500M Muse tokens/week); Maximum $100/month (3B tokens/week), per Meta's subscription help (read 28 October 2026).

Dots vs Muse vs Instinct: which fits your situation?

You already use ChatGPT and want connected context

Dots supports shared ChatGPT memory and existing plugins in the reviewed docs. Check the Plugins tab rather than assuming a fresh assistant starts with no access. Turning off Memory only stops future sharing. If you want to remove information already held, OpenAI says deleting your dot starts deletion of its context, while ChatGPT memories and files created elsewhere need separate attention.

You want an assistant reached by ordinary text or call

Instinct's homepage explicitly describes those channels and everyday tasks such as groceries, appointments and travel. Its terms are less specific than OpenAI's FAQ about which action needs which approval. Before delegating consequential work, establish boundaries and check the connected service's own security and sharing settings. The terms warn that actions may not be reversible.

You want reminders, side chats and a file workspace

Muse's help page documents all of these: reminders in Upcoming, separate side chats, and finished artifacts in Library. It names PDFs, spreadsheets, documents, images and podcasts. It also tells users to monitor for inaccuracies or unexpected actions. If training or retention is your deciding factor, Meta's privacy help now documents a default-on training toggle with a retroactive opt-out; review it before connecting sensitive accounts.

You administer an Enterprise workspace

Dots has the most specific workspace guidance among the sources compared here. OpenAI says access is off by default during the 4-18 November admin preview, then available by default unless an admin opts out, with separate permissions for local computers, cloud browser use, network access and a password manager. Enabling dots does not grant every app. Review connected-app authorization and all assigned roles too: one role granting access can leave it enabled even if another has it turned off. Enterprise phone messaging through iMessage, RCS or WhatsApp is unavailable in the reviewed docs. Admins can pause personal dots and workspace owners can delete them, but separate Work/Codex tasks and independently scheduled work may continue. Removing access is not proof that every task stopped. UAE data or inference residency is excluded. Check effective roles before and after 18 November; enabling access does not create a dot automatically.

Before connecting Muse: review both permission defaults

Meta's current help separates connector access from web access. In Settings > Permissions, connector Ask for some actions means approval before every write and important read; Always ask covers every connector action. Website defaults are a separate choice: Ask for some actions covers possible information sharing or unfamiliar websites, while Always ask covers every website. A strict connector setting does not establish the same website setting.

Check the actual connected-account list too. Many connectors can be retrieve-only, but Facebook, Instagram and Threads may already be connected through a shared Accounts Center. Meta says it does not review custom connectors or how they use your information. Review saved approval scopes rather than treating a connection as permission for every future action.

Disconnection stops data exchange, not necessarily retained memories or chat history. Training controls, deletion and scheduled work are separate checks. Meta's reset permanently deletes chat history, files and active tasks and cannot be undone; do not use a reset casually as a substitute for disconnecting one service.

Current Meta help re-read 8 November 2026: guidance and approvals, connectors and data controls. See our Muse permissions guide for the wider review. These are documented controls, not an account test or a finding that Muse is safer than the alternatives.

What changes for workplace buyers on 18 November?

OpenAI's 4 November release notes and current workspace guide describe an admin preview for eligible Enterprise, Edu and Healthcare workspaces until 18 November 2026. Access is off by default during that preview; it becomes available by default unless the admin opts out. The getting-started page separately says Enterprise preview is free through 18 November and billing then begins for enabled dots. Do not read this as a universal free plan or assume delegated Work/Codex usage is included.

Before that date, review effective member roles, enabled personal dots, their saved cost tiers and independent tasks. Pausing a personal dot does not delete its data or guarantee every separate task stops. Dots access, app authorization, browser/network controls, data residency and billing are separate checks. We have not tested these controls.

Source conflict: the older workspace URL still serves beta/off-by-default guidance. We use the renamed current workspace guide, getting-started page and dated 4 November release notes for this transition. This is a documentation update, not a rollout test.

Common questions

Is disconnecting an account enough to delete its data?

No, according to both Instinct's terms and OpenAI's dots FAQ. Instinct points to Workspace for indexed-input deletion requests. OpenAI says disconnecting stops new access but does not clear context; deleting the dot starts deletion of that context, not separately stored files, ChatGPT conversations or ChatGPT memories. Meta's privacy help separately describes message/chat/file deletion, full reset and best-effort forgetting. Do not confuse deleting source content with removing learned memories.

Can I edit what Dots remembers about me?

OpenAI says individual dot memories cannot currently be viewed, deleted or directly modified. ChatGPT Memory has separate controls, and deleting the dot starts deletion of its own context. Those are different operations.

Do these agents use my information for model training?

Do not treat them as having one shared policy. Instinct's terms and privacy policy contain both an opt-out and specific exclusions, including Vault and directly received Google Workspace API data. OpenAI distinguishes personal plans from Business, Enterprise and Edu, and personal plans use a ChatGPT setting. Meta's privacy help documents a training toggle that is on by default with a retroactive opt-out.

Does Dots ask before every action?

No blanket promise is made in the reviewed FAQ. OpenAI distinguishes actions requiring takeover, actions that may need approval each time, and supported actions with scoped advance approval. Custom Rules cannot override core safeguards. Its workspace guide also says disabling Custom Rules does not make every action require approval.

Which is cheapest or most reliable?

This evidence does not support either ranking. Muse documents a free weekly allowance plus Power $20/month and Maximum $100/month tiers. OpenAI documents Pro/Business Premium access and a separate eligible-Enterprise preview and billing transition. Exact Enterprise dollar rates were not established in these sources. Instinct's current plan prices could not be verified from the public sources checked. We have not run a comparative task test, measured completion rates or checked real signup eligibility for each region.

Before you connect accounts

  1. Confirm your plan, region, price and usage allowance.
  2. Connect only the accounts needed for one low-risk task.
  3. State what needs approval, what is forbidden and who may receive information.
  4. Check training settings before sharing sensitive material.
  5. Write down the separate steps for disconnecting, deleting retained context and deleting files elsewhere.

For another text-based option, see Poke vs Instinct. For broader selection, read personal assistant agents compared.

Sources and scope

Dots and Muse sources re-read 5 November 2026. This is a comparison of public vendor documentation, not an independent security audit, hands-on review or proof of signup availability. A missing detail means it was not established by these pages, not that the feature or policy does not exist.

See our methodology or report a correction.