Vercel Labs has officially unveiled json-render – a groundbreaking open-source framework designed to formalize the emerging concept of “Generative UI.” Launched in January 2026 under the Apache 2.0 license, the project has quickly become a standout on GitHub, amassing over 13,000 stars and 200 releases. The technology allows artificial intelligence models to construct complex, structured user interfaces using natural language prompts.
The core philosophy of the framework is to give developers absolute control over AI – driven outputs. By defining a catalog of permitted actions and components via Zod schemas, the system ensures that any JSON specification generated by a Large Language Model is strictly constrained to that catalog. This prevents the AI from generating unsupported elements or unpredictable code.
Once the AI generates a flat JSON tree of typed elements, the Renderer component maps these references to real implementations. Vercel CEO Guillermo Rauch told The New Stack that the approach “plugs the AI directly into the rendering layer” and described it as a “very disruptive technology.” To ensure a smooth experience, the framework renders output progressively as the model streams its response.
The framework is highly versatile and ships with 36 pre-built shadcn/ui components for teams seeking a quick start. It includes specific packages for PDF generation, HTML emails, video via Remotion, and 3D scenes through React Three Fiber. Additionally, renderers are currently available for multiple libraries:
- React
- Vue
- Svelte
- Solid
- React Native
Community feedback on platforms like Hacker News has been diverse, with some developers noting that providing AI with an accessible structure like this feels familiar. One user compared the new framework to the development tools of the late 90s.
Giving an AI accessible structure to this gets AI into the realm of the various 4GLs back in late 90s which made user created forms so much easier.
Practical success stories are already appearing within the developer community. One early adopter mentioned significant progress in building text – to – dashboard features, noting that this framework proved to be more robust than previous attempts using standard structured output APIs and older models from the GPT – 4 era.
However, the project has also faced skepticism from those who question the necessity of a new system. Some critics suggested that existing standards like OpenAPI and JSON Schema could already describe the required data structures without needing a specialized framework.
OpenAPI, JsonSchema, GraphQL all describe Data.
This describes User Interfaces. The closest alternative would be to just return React JS code directly.
But this adds a layer of constraint and control, preventing LLMs to generate e.g. malicious React JS code.
In response to these concerns, advocates pointed out that while tools like GraphQL are designed for data, json-render focuses specifically on user interfaces. This adds a crucial layer of safety, preventing models from generating potentially malicious React JS code by keeping them within a controlled sandbox.
The shift toward this new way of working was also highlighted on Reddit. A developer there observed that the industry has been moving toward constraint – based systems for years, citing design tokens and Storybook as precursors. This transition simply pushes that boundary further, moving from authoring components at build – time to composing them dynamically at runtime.
Google has also entered the space with a comparable project called A2UI – or Agent – to – User Interface – which was quietly released in late 2025. While the two projects share a similar high – level pipeline, they are designed to solve different technological challenges. An analysis on Medium noted that while json-render is a tool tightly coupled to specific application components, A2UI positions itself as a protocol for cross – agent interoperability.
Currently maintained by Vercel Labs, json-render is written in TypeScript and uses a monorepo structure managed with pnpm. The project remains accessible to the public via npm under the @json-render scope, with full documentation that includes a quick start guide, a dedicated playground for testing, and specific examples for each supported renderer.
