Replit: our review

Replit combines a browser-accessible code editor with an agent — Replit Agent — capable of building an application end to end: writing the code, running it, creating a database, and publishing the result. The difference from a simple generator lies in the persistence of the environment: the project remains a full development space, not just a disposable preview, with a terminal, installed dependencies and a real database.

What it actually does

You describe a project to the agent — a booking application, an internal tracking tool — and Replit Agent plans the steps, writes the code, installs the necessary packages and launches the application in the environment. Each validated step creates a save point called a checkpoint, which captures the project's full state: files, configuration, database contents and conversation context. You can go back to any earlier checkpoint, or move forward again if the rollback went too far — a flexibility that neither Lovable nor Bolt.new documents in this much detail.

Getting the project out, and the licensing question

There are two ways to get a project out of Replit: download it as a zip from the file explorer, available on all plans, or sync it with a GitHub repository via the built-in Git tool, which works both ways and re-syncs easily after a connection drop. One point deserves particular attention, absent from the marketing pages but explicit in the terms of service: publishing an application with public access automatically places it under the MIT licence, which lets anyone view, copy, modify and redistribute it, on the platform or off it. For a firm testing an internal tool without realising it has made it public, the consequence is direct: the code stops being exclusive. Keeping the application private avoids this sharing, but requires knowing this from the start.

What an iteration costs

Replit bills the agent by effort rather than by message: a simple fix costs less than a full feature, grouped into a single billed checkpoint. The point worth remembering, spelled out plainly in the documentation: every interaction is billed, including when the agent replies with text without changing the code, or when the result doesn't match the request. Rolling back restores the project, not the credit already spent getting there. The Core plan, at $20 a month billed annually, includes $20 of monthly credits; the Pro plan, at $90 on annual billing, includes $100. Beyond this allowance, hosting published applications separately bills outbound data volume, compute time and number of requests — three distinct meters that add to the bill of a project receiving real traffic.

Where it falls short

Two reservations worth knowing. First, the automatic MIT licence on any public application: it's a platform choice consistent with its open-source culture, but one that surprises a company used to "publishing" not meaning "giving away." Second, effort-based billing makes it hard to estimate a monthly budget in advance: unlike a per-user rate, the cost depends on the number of back-and- forths with the agent, their complexity, and the traffic received by the application once published — three variables that only real usage lets you measure.

Our recommendation

Replit suits a technical team that wants to combine development and hosting in one place, and that systematically checks each project's visibility status before publishing. See also our selection of app generation tools. Best avoided if the agent needs to run unsupervised on repeated tasks: every exchange adds to the bill, whether it succeeds or not, and code published with public access doesn't stay exclusive.