Replit: what it is, how Replit Agent works, pricing, and limits

Replit in three lines

Replit is an online platform where an AI agent, called Agent, builds a web or mobile app from your requests, with the database, publishing, and version history in the same place. It is aimed at people who do not code as well as people who want to stay in control of the code. It bills a dollar allowance of credits that the agent draws down by effort, so you cannot work out a number of apps from it.

This page is an analysis of Replit’s public pricing, documentation, and terms, as recorded on the date shown. No app was built to write it. For the concept itself, see the guide what is vibe coding.

What Replit actually does

According to the pricing page and the documentation:

• App type. Web and mobile apps (Expo); submission to Apple through its beta service, Google Play not yet supported. source Expo is a toolkit for building mobile apps; you pick the “Mobile app” type when you start the project.
• Data. Managed PostgreSQL and built-in object storage; authentication through Replit Auth or Clerk. Development databases are not billed; production databases are billed on compute time and storage.
• User login. Replit Auth shows a Replit-branded sign-in page where your users sign in with a Replit account: the documentation points it at prototypes and internal tools. For a product under your own brand, it points to Clerk.
• Publishing. The app goes live on a .replit.app address. You add a custom domain in the publishing settings, and you can buy a domain inside Replit from a personal account, with no fees added by Replit.
• Code. Built-in Git, GitHub sync from the Git pane, branches managed in the editor, import of a GitHub repository. source You can also connect a GitHub or GitLab repository.
• A limit on every plan. 20 Replit apps running at the same time.

What Replit bills: a dollar allowance

At Replit, a credit is not an action: it is money. Your subscription holds a monthly dollar amount, and everything you do draws on it: the agent’s work, but also hosting for published apps, the production database, and storage.

Why you cannot predict a number of apps

Four reasons, all from the documentation:

1. The agent is billed by effort. The longer it works on a request, the more that request costs. You read the price in the Agent tab once the checkpoint exists, not before.
2. No unit price is published. The vendor publishes no price per checkpoint. There is no public figure to multiply by a number of requests.
3. The same allowance pays for running the app. Going by the publishing rates listed in the notes at the bottom of this page, the base fee of an “Autoscale” deployment equals one tenth of the Core plan’s monthly credits, before compute and requests. A shared reserved VM takes three quarters of them. A dedicated reserved VM costs more than Core’s monthly credits on its own. Every live app shrinks what is left for building.
4. The mode changes everything. In “Free Mode”, the agent uses no credits within an allowance; in “Power” and “Max” modes, it does. Some features also call other providers’ APIs, billed at the provider’s public rate and deducted from your credits.

What you can control: a confirmation is requested before any paid agent action, alerts and a budget cap can be set in the account, and “Plan” mode reduces consumption (Replit documentation, “Managing spend” (accessed November 11, 2026)).

The plans in brief

The free plan is called Starter. It is described in the documentation but missing from the pricing page, and its quotas are not given. The Core plan costs $20 per month, or $18 per month billed annually, and includes 20 dollars of credits per month. The Pro plan starts at $100 per month, with a credit amount you pick from several tiers. Enterprise has custom pricing.

The plan table and the tiers are further down; the Replit pricing page covers what the free plan allows and the math on the tiers. On the number of collaborators on Pro, the pricing page contradicts itself: the plan card says up to 15 collaborators, while the detailed table on the same page says “All team members.”

Credit rollover: the vendor contradicts itself

The vendor’s pages contradict each other on this point. The pricing page says credits reset each month. The Pro plan documentation says unused credits roll over for two months (Replit documentation, Pro plan (accessed November 11, 2026)). No rollover is mentioned for the Core plan.

Until the vendor settles it, the cautious move is to plan on a monthly reset and size your plan for an ordinary month, not a peak one. Credit packs bought on top are valid for six months.

Hosting and data

Where your data is hosted was not recorded: the pages read do not say. What is documented: object storage (“App Storage”) is backed by Google Cloud Storage; a production database can reach 100 GiB on a paid plan, 1 GiB on Starter; the Pro plan lets you restore the database up to 28 days back. The Enterprise plan lists single-tenant and private networking options. Self-hosting was not recorded.

On ownership, the terms of service say you keep your rights in the content you submit. They do not deal separately with AI-generated code.

The agent and its three modes

• “Free Mode”: everyday work, without credits, within an allowance that renews every five hours on the Core and Pro plans. Replit picks the AI model. On Core, the pricing page lists up to 30 hours of chat and 60 projects in this mode.
• “Power” and “Max”: paid from your monthly credits, with a choice of model.
• On the Starter plan: the agent is limited to “Lite build”, “Free Mode” has a daily cap, and “Power” and “Max” are not available.

The documentation adds that all subscribers receive free checkpoints, granted one time only.

Our take, based on the documents

• If you want a fixed cost, Replit is not built for that. The budget cap limits spending, but once it is reached, usage-billed services are suspended until you raise it or the next cycle starts.
• If keeping the code matters, built-in Git and GitHub sync are documented with no plan condition recorded.
• If you are going mobile, the documented path runs through Expo and Apple; Google Play is not yet supported.
• If you are new to the tool, Starter lets you watch the agent work, but the published app goes offline after 30 days and carries a Replit badge.

Alternatives to Replit

The table below compares Replit with Lovable, Bolt, and Base44 on the same rows. To go further:

• Replit vs Lovable: two ways of counting, a dollar allowance versus credits per message.
• Base44 vs Replit: Base44 publishes an estimated cost per type of action, Replit does not.
• Replit, Base44, and Lovable alternatives and AI app builders compared, for the wider picture.