How does AI work in Draftbit?

You work with the agent in the AI Chat tab of the Builder. Each conversation is a thread. When you start one you pick the agent, Claude Code from Anthropic or Codex from OpenAI, and the model it should use. Defaults for both live in Agent Customization, next to a CLAUDE.md or AGENTS.md file you can edit to tell the agent about your project’s conventions and important files.

The agent runs inside the same sandbox as your app’s preview server, so it works on the real project source rather than a copy, whether that is a React Native and Expo app, a Vite web app, or an Astro website. Beyond editing code, it has tools to type check and lint the project, add or remove npm packages, check the preview server’s status and logs or restart it, manage environment variables, call REST and GraphQL endpoints you have connected, and search the Draftbit docs. When it needs more information, or approval for a sensitive step, it pauses the run and asks you in the chat.

You can add context by selecting a component on the canvas, referencing a file, attaching an image, or choosing a Design Block. Edits from the agent appear in the same project the visual editor and code editor use. You save the result as a save point and can roll back to an earlier one if a change goes wrong.

Agent usage is paid with Draftbit credits, or you can connect your own OpenAI, Anthropic, or OpenRouter key, or a ChatGPT or Claude subscription. Threads on your own account skip credits entirely, on every plan including Free.