AgentComparison
Replit · product · application building agent

Replit Agent

Builds and fixes whole apps from plain-language instructions inside Replit.

Group: Coding agents

Replit documents an agent that creates applications, sets up projects, checks its work and fixes problems from natural-language instructions.

Hands-on pilot completedChecked 22 October 2026Vendor documentation plus a small hands-on pilot

The decision in plain English

Separate building an app from maintaining a production app; record deployment and usage charges before recommending a plan.

Best fit

Building apps and related artifacts from natural-language requests.

Poor fit

Assuming a prototype is production-ready without review.

Documented capabilities

Vendor statements, not verified task outcomes.

  • Writes code, sets up infrastructure and tests Source · checked 2026-10-22
  • Plan mode before changes Source · checked 2026-10-22
  • Checkpoints and rollback Source · checked 2026-10-22

Permissions and human control

  • 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.

Documented controls do not prove that an agent always follows them.

Limits and open questions

  • Generated work still needs code, deployment and security review.
  • On Starter the usage dashboard requires an active subscription, so free-tier consumption cannot be self-monitored (observed 2 November 2026).
  • 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.

Privacy and 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.

Feature-by-feature: what is documented and what is not

Official sources checked 22 October 2026. A documented feature is a vendor statement, not proof of success.

FeatureEvidence statusDetail and limit
Plan before changesDocumentedPlan Mode maps out work before code/data changes. Paid-path planning can bill even when no code changes. Source
Free versus paid modesAllowance and approval matterFree 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. Source
Starter capabilitiesLimitedCurrent billing table lists daily-capped chat/Free Mode, but no full build, Plan Mode, connectors or background tasks on Starter. Source
Artifacts and appsDocumentedWeb/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. Source
CheckpointsReviewable rollbackCompleted requests create checkpoints. Vendor says these support rollback; verify what is restored rather than assuming an external payment or message is undone. Source
Background capacityPlan-dependentBilling docs list 0 active tasks on Starter, 1 on Core and 10 on Pro. Source
Paid effort and APIsVariable costPaid checkpoints reflect request complexity. Third-party model services can deduct public API rates from credits; no fixed per-request total is claimed. Source
Budget visibilitySet caps before paid workCredits also cover cloud services. Alerts and hard budget caps are documented; usage dashboard data may lag up to 30 minutes. Source

How to start with Replit Agent narrowly

  1. Start with Free Mode and check the actual allowance in Settings > Usage.
  2. Describe one small task and review a plan before allowing code/data changes.
  3. Review the diff, test output and checkpoint after completion; do not assume rollback reverses external actions.
  4. Before Power/Max or deployment, inspect paid-action confirmation, cloud/model costs and billing caps.

Official product/setup source. These are documentation-based steps, not a hands-on setup test.

Hands-on pilot: three sequential prompts on one app, free Starter plan

Hands-on pilot, 2 November 2026 · We ran this ourselves on a free account. Small pilot, not a full evaluation.

We ran one synthetic free Starter account through three sequential prompts on a single app (a tip calculator) on 2 November 2026, all in the Agent's Free mode, with no card attached. Sequential prompts on one app are not independent trials. Timings below are the Agent's own reported working time; each matched what we observed. Results were checked in the running app itself, not from the Agent's summary.

CheckResultNote
Build a one-page tip calculator from scratchPass, about 2 minutesExact prompt: "Create a simple one-page tip calculator web app: bill amount input, tip percentage buttons (10/15/20%), and it shows tip and total. Plain HTML/CSS/JS." $100 at 15% returned a $15.00 tip and $115.00 total; switching to 20% recomputed instantly.
Add a bill-splitting feature to the same appPass, about 1 minuteExact prompt: "Add a 'split between people' field: a number input for how many people are splitting, and show the amount each person pays." $100 at 20% split 4 ways returned $30.00 per person.
Add a PostgreSQL database with a last-10 history listPass, about 10 minutesExact prompt: "Add a PostgreSQL database that saves every calculation (bill, tip %, split count, totals) with a timestamp, and show a history list of the last 10 calculations on the page." Entries persisted after a full reload; with 12 calculations saved, exactly the 10 newest showed. The Agent hit dependency problems mid-run and fixed them itself.

What else we observed

A pilot, not a verdict: one synthetic account, one app, three sequential prompts, free Starter plan, 2 November 2026. This measures capability and time, not dollar cost - effort-based checkpoint pricing applies to paid plans this pilot did not touch. The result is a shared, development-stage app, not a production-ready one; generated work still needs review.

Common questions about Replit Agent

What is Replit Agent for?

Building apps and related artifacts from natural-language requests. The vendor describes it as: Replit documents an agent that creates applications, sets up projects, checks its work and fixes problems from natural-language instructions.

Who should look elsewhere?

Assuming a prototype is production-ready without review.

How much does Replit Agent cost?

Replit's pricing page (read 2 November 2026) publishes subscription prices: Core at $20/month ($18/month billed annually) including $20/month in credits toward more powerful models, and Pro at $100/month ($90/month billed annually) including $100/month in credits, with more collaborators and longer database rollback. Credit mechanics per the AI billing doc (read 1 November 2026): Core subscribers get monthly credits; Pro credits are tiered and leftovers roll over for two months; credits cover Agent and cloud services. Free Mode supports no-cost work within a plan allowance. Confirmed paid work in Power or Max uses effort-based pricing; no fixed per-request total is published (last full check 22 October 2026).

What can it access and what needs my approval?

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.

What happens to my data?

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.

Can I use Replit Agent from Slack safely?

Talking to Replit Agent from Slack is different from giving Agent a Slack connector to search or send messages. Current setup docs require a Replit team workspace with Slack enabled, an eligible verified account, and installation permissions. Each teammate links their own account; requests use that account's Replit access. A thread belongs to its starter: other channel members can read replies and files but cannot take over approvals. Use /replit unlink to disconnect your own account for that workspace. Removing the workspace bot is separate and does not remove saved personal links. We have not tested this integration.

Source · Source

Does Slack access include Cloud Agents for an existing repository?

Replit's 27 October announcement (updated 29 October) lists existing-repository Cloud Agents for Pro and Enterprise, producing GitHub pull requests or GitLab merge requests for engineering review. Its general Slack integration statement does not grant every plan that existing-repo entitlement. Current Slack setup docs additionally require the feature enabled in your Replit team workspace. Check both plan and workspace availability. Learning review feedback into team skills is early access; Teams support is described as coming soon. Vendor-described browser tests and videos are not our own test results.

Source · Source

Can Replit Agent buy supplies, and does connecting Link approve a purchase?

Replit announced Agentic Commerce on 12 November 2026 for Core and Pro, starting in the US. It describes a browser that compares products and prepares checkout, then asks for approval of the final total through Link. Connecting Link or taking over a merchant sign-in does not approve payment. The purchase guide says changed terms need new approval, and payment confirmation is separate from merchant order confirmation. If either status is uncertain, do not repeat the purchase. There is a scope mismatch to check before buying: the launch announcement says one-time purchases only, while the current guide also describes fixed-price subscriptions with identical initial and renewal totals. Do not assume recurring purchases are enabled for your account. The guide excludes trials, introductory prices, usage-based billing, prorations and variable renewals; ending payment authorization is not merchant cancellation. Sources read 12 November 2026; we have not tested shopping or payment controls.

Source · Source

Can Replit run Google Ads, and which account pays?

The current guide documents Performance Max campaigns, with campaign activation approved separately from account connection. For a new Replit-managed account it requires Pro, a personal Workspace, available Replit credits and auto-reload. Replit creates and manages that account, deducts ad spend from the shared credit balance, and can replenish credits by charging the saved payment method. An existing connected Google Ads account keeps its separate Google Ads billing arrangement. Campaign creation requires an average daily budget of at least $15 USD, within the setup limit. Average daily budget is not a hard daily cap, and a Mission budget is not an advertising spending cap. Preparing and managing ads incurs Replit usage separately from ad spend. Reporting and charges can arrive after activity, including after a pause. Confirm the stop condition rather than assuming an automatic stop. Source read 12 November 2026; not an ad-buying test.

Source

Does an enabled Replit ad campaign mean ads are serving or sales are proven?

No. Replit's guide says enabled is a saved campaign status, not confirmation that Google has approved or begun serving the ads. Account eligibility and Google's review still apply. The documented launch supports Performance Max; Replit-created accounts can run one campaign at a time, and the Replit-managed launch is for personal rather than Enterprise Workspaces. Google promotional credits are not included. Conversion tracking, reporting routines and scheduled actions need their own setup. These are vendor-documented capabilities, not our evidence of leads, sales or return on ad spend. Our free Starter build-and-edit pilot did not test commerce, advertising, auto-reload or payment approval. Source read 12 November 2026.

Source

Has AgentComparison tested it?

Yes, a small pilot. On 2 November 2026 we ran three sequential prompts on one app from a synthetic free Starter account: build a tip calculator, add bill splitting, then add a database-backed history. All three passed functional checks at no observable charge. That measures capability and time on the free plan, not paid-plan cost. Details in the hands-on section above.

How to evaluate Replit Agent yourself

  1. Pick one low-risk task that matches its documented capabilities: writes code, sets up infrastructure and tests.
  2. Write the expected output and what counts as failure before you start.
  3. Connect the minimum access needed, and check approval settings first.
  4. Run it twice and compare, then review every action it took.
  5. Check cost after any free allowance. Unknown price is not free.

Related

Documented use cases: Software development and repository work.

Sources and history

Change history

Hands-on pilot results reported above with their limits; everything outside that section remains vendor documentation.

Report a correction