Replit Agent turns a plain-language description into a working application inside a Replit workspace. You describe what you want, optionally pick an output type (web app, mobile app, slides, or a design), and the agent writes the code, provisions infrastructure, and tests the result. You then refine it through further chat messages and publish from the same project.
The agent searches the web when it needs current information, tests its own work as it goes, and records checkpoints so you can roll back to an earlier state. For open-ended requests such as performance work or design overhauls, you can switch on extended thinking and higher-capacity models. Built-in services cover databases, authentication, secrets, and third-party integrations, and the agent can query connected sources such as BigQuery, Linear, Slack, and Notion from within the chat.
Usage runs in tiers: a free allowance for Core and Pro subscribers with periodic resets, a Power mode, and a Max mode that uses frontier models with an optional Turbo setting.
Features
- Chat-driven builds: describe the app, iterate with feedback, and publish without writing code
- Multiple outputs per project: web apps, mobile apps, data dashboards, slides, and files such as CSV, PDF, PowerPoint, and Markdown
- Design Canvas: visual designs and prototypes as a separate output type
- Self-testing and checkpoints: the agent verifies its changes and creates restore points for rollback
- Extended thinking and high-power models: optional settings for larger or more ambiguous tasks
- Built-in services: database, auth, and secrets alongside third-party integrations
- Connected data sources: query BigQuery, Linear, Slack, and Notion directly from the conversation
- Web search: the agent looks up current information while building
