Vercel Labs · 2026-09-18 · notable
json-render 0.21.0 — Vercel Labs adds a TanStack Start renderer
json-render, Vercel Labs' framework for turning model-generated JSON into real UI components, ships 0.21.0. The release adds a TanStack Start renderer for whole routed apps with server rendering, plus experimental spec composition.
Vercel Labs' generative UI framework now renders whole routed applications, not just single views.
Key specs
| GitHub stars | 17,202 |
|---|
What is it?
Version 0.21.0 of json-render adds @json-render/tanstack-start, a renderer that turns a JSON spec into a full application with file-based routes, reusable layouts, SSR loaders, head metadata, prerender paths, client navigation and route fallbacks. json-render itself is the open-source framework Vercel Labs builds for generative UI, where a model writes JSON and your own components render it.
How does it work?
The pipeline is AI, then json-render, then UI. A developer calls defineCatalog to declare the components and actions a model is allowed to use, and defineRegistry maps each catalog entry to a real implementation. The model can only emit JSON that fits that catalog, so a bad generation fails validation instead of rendering something unexpected. This release also adds experimental_composeSpec and experimental_createEvaluator, which assemble a validated spec from candidate pieces the app already owns.
Why does it matter?
Letting a model write free-form component code is hard to ship safely; pinning it to a fixed catalog is what makes a generated interface predictable. Until now the framework produced views you dropped into an app you had already built. The TanStack Start renderer moves that line out to routing and server rendering, so a single spec can describe the application itself. Vue registries also gained catalog-declared named slots in 0.21.0, and streamed React renders were stabilized.
Who is it for?
web developers building LLM-driven interfaces
Try it
npm install @json-render/core @json-render/react