Replit Agent

Replit Agent is the AI builder inside Replit's cloud IDE: you describe an app, it plans, writes, runs and publishes it. Replit lists Core at $20 a month and Pro at $100, and bills Agent work by effort, one checkpoint at a time. A Starter plan gives daily credits and one free app that expires after 30 days. Checked 31 October 2026.

At a glance

What it is

Replit's site frames the product as three starting points: Chat (ask questions, research and plan), Build (create and publish apps and sites) and Design (create interfaces and content). Agent is the builder behind the Build path. You describe what you want, Agent writes the code in a project that lives in Replit's cloud, runs it, and lets you publish it.

The whole environment is hosted, so there is nothing to install: an editor, a runtime, a database, storage and hosting sit in the same account. That is the difference from Lovable and Bolt, which are app builders on top of other hosts, and from Claude Code or Cursor, where you bring your own machine. Replit is closer to a cloud IDE with an agent attached.

How Agent works with you

The docs describe Agent as "Replit's AI builder." It can plan changes, write code, explain behaviour, debug issues and improve your app. Three features shape the experience.

• Plan mode. Turn on Plan in the prompt composer and Agent thinks before changing files and waits for your approval. The plan lists what and why, "Done looks like" criteria, out-of-scope items and numbered build steps.
• Checkpoints. Each request produces a checkpoint. Open History in the Agent panel to see what changed, with "Rollback here" and "Changes" controls. A rollback restores files, Agent memory, tasks and, if you tick the box, the database.
• Preview. The docs say Agent may summarise changes afterwards, and you should test in Preview yourself.

That last line is good advice for every builder in this category.

Agent modes

From the agent modes page:

Economy Mode is no longer available; Power Mode replaces it at the same price. Routines run only in Power or Max Mode and need a per-run budget. On Pro, you can run up to 10 Agent tasks in parallel.

Plans and prices

From replit.com/pricing and the plan docs on 31 October 2026. Prices are subject to tax by location.

The Pro page also shows four larger monthly amounts ($215, $425, $825 and $2,000) without labels. The docs say Pro credits come in tiers with bigger discounts at larger sizes, so these are likely credit tiers. Replit does not label them, so treat that reading as ours.

Starter's restrictions matter. Full build, Plan Mode, third-party connectors, Replit AI Integrations, converting a Design Canvas design into an app with backend functions and artifact types other than web and mobile apps all need Core or Pro.

Core or Pro

The docs make the split simple.

A solo builder who ships one or two apps rarely needs Pro. A small agency running several client builds at the same time does. Whichever you pick, test with Starter first and watch which restriction you actually hit.

How the billing works

Replit bills Agent work by effort, not by the message. The AI billing page says:

• Charges depend on the complexity of a request. A simple fix costs less than a full build, and a full build is bundled into a single checkpoint.
• Each request produces one checkpoint once Agent finishes the work.
• "Replit asks for confirmation before a paid action starts."
• On the paid path, Plan Mode can incur charges even when it does not change code.
• Calls to third-party services such as Claude, ChatGPT and Gemini are billed at the provider's public API rate, deducted from your credits.
• All subscribers get one-time free Agent checkpoints. Core gets monthly credits; Pro gets tiered monthly credits that roll over for two months.
• Credits also pay for published apps, storage and databases.

Managing spend: hover over the usage icon in the Agent tab to see per-checkpoint cost, use the usage dashboard (data can take up to 30 minutes to appear), and set alerts and budgets on the Billing page. Larger credit packs get discounts. Replit also offers an interactive estimator at agent-pricing.replit.app. The page does not give dollar rates per checkpoint.

The practical effect: your bill follows what you ask for, not how many prompts you send. A vague, sprawling request costs more than a small, well-aimed one.

Privacy and data

Replit's model improvement page says Replit uses content you provide, including requests, instructions, surrounding context and generated outputs, to develop and improve its models and AI features, by default for Pro users unless they opt out. To opt out, go to Settings, Personalization and turn off Model Improvement. Opting out means new content tied to your account will not be used, and it does not delete existing content. For Enterprise, "Replit does not use customer content to train its models." The page does not address Starter or Core.

Replit may still process content to run, secure and troubleshoot the service, enforce its terms and comply with law, whatever the setting.

For code that belongs to a client or an employer, do not use a plan whose terms you have not read.

Publishing and databases

You publish from the same workspace. Starter allows one free published app that goes offline after 30 days; Core removes that limit. Databases and object storage run inside Replit, and their usage draws on credits. The docs have separate pages for publishing costs and for importing a project from GitHub. We did not read the deployment cost tables. See docs.replit.com/billing/deployment-pricing for those.

If you already have a repository, Replit can import from a Git provider, and the docs cover connecting a repo and managing access. That makes it possible to move in and, with Git, out.

What it does well

• One place. Editor, Agent, database and hosting in a single account.
• Plan then build. A real planning step with approval and "done looks like" criteria.
• Checkpoints and rollback, including the database if you choose.
• Cross-device. Web, desktop and mobile apps.
• Clear modes. Free for routine work, Power and Max for harder tasks.
• Parallel tasks on Pro.

Limits and cautions

• Unpredictable cost. Effort-based billing is fair but hard to forecast; use budgets.
• Plan restrictions. Full build and Plan Mode need a paid plan.
• Free apps expire. The Starter app goes offline after 30 days.
• Hosted lock-in. The code is yours, but the environment and database live at Replit; plan an export path.
• Data default on Pro. Opt out if your work is sensitive.
• Tests are on you. The docs tell you to test in Preview.

Replit versus the others

Also see best vibe coding tools and vibe coding vs agentic coding.

Who should pick it

Replit suits beginners and solo builders who want everything in one tab, and teams (up to 15 builders on Pro) who share pooled credits. It suits less if you need your own cloud account for compliance, or if you want the editor on your own machine. If you already know how to code, an editor agent plus any host gives you more control; if you do not, Replit's all-in-one approach removes many decisions.

Planning spend on Replit

Effort-based billing rewards a few habits. Use Plan mode to agree the work before paying for a build. Keep requests small and specific, because a full build is bundled into one checkpoint and a simple fix costs less. Use Free Mode for routine edits and keep Power or Max for harder problems. Set a budget and an alert before the first heavy session, and check the per-checkpoint cost by hovering over the usage icon. After a week, you will have a feel for typical cost per checkpoint, which the public pages do not give.

Habits that keep Replit projects tidy

• Write the goal and "done looks like" in your first message, as the plan view does.
• Roll back with History instead of piling corrections on a bad checkpoint.
• Keep secrets in the secrets store, not in code or chat.
• Test in Preview after every checkpoint, as the docs advise.
• Review the usage dashboard weekly, remembering it can lag by up to 30 minutes.

Our take

This is an editorial reading of Replit's pricing and documentation, not a hands-on test.

The plan docs are clear about what each tier unlocks, and the billing page is honest that costs follow effort. That honesty is also the weakness: without a rate card you cannot know a monthly bill in advance. The right response is to use the budget and alert tools from day one, and to learn what your typical checkpoint costs from a first week on the free tier.

Treat Replit as a good place to build and host a first version, with Git as your exit. If you plan to keep the app for years, connect a repository early, keep secrets in your own store, and decide on the data-training setting before you paste in anything that belongs to a client.

Checklist

• Start on Starter and note your daily credit use
• Use Plan mode for anything bigger than a small fix
• Set a budget and alert in Billing before heavy use
• Opt out of Model Improvement on Pro if the work is sensitive
• Connect a Git repository early
• Test every change in Preview
• Re-check pricing on replit.com the day you subscribe