For AI agents: the complete documentation index is available at /llms.txt, the full documentation bundle is available at /llms-full.txt, and this page is available as Markdown at /guide/genui/reactlynx/overview.md.
This is the dev preview website. Check out the document at lynx-stack.dev

ReactLynx overview and architecture

The model authors a ReactLynx function component and plain CSS. The host owns the entry point, compiler configuration, runtime dependencies, and bundle loading. Unlike A2UI Catalogs or OpenUI Libraries, this integration uses generated source code and requires compilation before rendering.

Generation flow

  1. Send REACTLYNX_SYSTEM_PROMPT and the user's request to your model.
  2. Accumulate response deltas and parse the complete response with parseReactLynxSource.
  3. Call buildReactLynx on Node.js with an AbortSignal and a progress callback.
  4. Publish all emitted assets, preserving their relative paths.
  5. Load the corresponding Web or Native entry bundle in your Lynx host.

The host supplies root.render(<App />) and imports App.css. Generated code imports only @lynx-js/react. It uses Lynx elements, React state, and Lynx event handlers to implement interactions locally within the page.

Application responsibilities

Keep model credentials and storage access on your backend. Connect cancellation to the build's AbortSignal, show compilation diagnostics, and update the preview only after a completed build has been published. For follow-up edits, send the previous source with the request and ask for both complete files again.

The compiler uses a separate child process with a restricted environment and bounded resources. It validates source syntax and import requests before compiling, and executes only host-owned build configuration. These checks do not turn generated code into a component allowlist or a general runtime sandbox; your host controls where generated bundles can run.

See source and builds and system prompts.