json-render is an open-source generative UI framework from Vercel Labs, with more than 18,700 stars on GitHub. It solves the central problem of AI-generated interfaces — unpredictability — by letting the model produce a JSON spec that can only reference components and actions from a catalog you define. The result streams into safe, predictable, on-brand interfaces across web, mobile, documents, email, video, and even terminals. The project is very actively developed, with frequent releases through 2026.
A json-render app has three parts: a catalog that declares your components, their props as Zod schemas, and the actions they may trigger; a registry that maps those names to real implementations; and a renderer that turns the model’s streamed JSON into live UI. Because the AI works inside guardrails, output is schema-valid every time, while progressive streaming renders the interface as tokens arrive. One catalog can target many renderers — React, Vue, Svelte, Solid, React Native, Next.js, Remotion video, React PDF, React Email, Ink terminals, images, and Three.js 3D scenes — and 36 pre-built shadcn/ui components make it easy to start.
json-render is completely free and open source under the Apache-2.0 license. You pay only for the LLM calls your application makes.
json-render is for product engineers building AI features that generate dashboards, reports, forms, and personalized interfaces — anywhere free-form AI output is too risky to ship. Because one catalog drives many renderers, it also helps teams reuse the same AI-generated experiences across web, mobile, documents, and email from a single source of truth.