AI-generated interfaces still using JSON? OpenUI wants to replace this intermediate representation

AI-generated interfaces usually follow one path: the model outputs JSON, the frontend parses it, and then maps it to components. It works, but token overhead is high, and streaming rendering is awkward. OpenUI wants to replace this intermediate representation.

It is an open-source generative UI framework that supports React, Vue, Svelte, and Angular. At its core is a streaming-first language designed specifically for AI-generated interfaces. According to the project, it uses up to 67% fewer tokens than JSON. The React version also includes a ready-made component library and chat interface, shortening the integration path.

The key here is not just saving tokens. JSON was not designed for streaming UI. When the model outputs token by token, it is difficult for the frontend to render while receiving. OpenUI consolidates the contract between model output and UI components, bringing the generation process closer to the interface itself.

Some commenters think treating JSON as the baseline is already generous. That judgment is a bit harsh, but it points out the problem: using JSON as the default format for AI-generated interfaces has always been a compromise. What OpenUI wants to do is let models directly output an intermediate representation that can drive the UI, reducing parsing wait times and frontend-backend alignment costs.

For developers, the React version is the first place to look at the component library and chat interface; other frameworks can be integrated according to the repository instructions. For ordinary users, the more direct change is shorter waits and a smoother process when generating interfaces. Whether it can become a standard depends on the ecosystem and cross-framework adaptation, especially outside the React camp.

GitHub repository: https://github.com/thesysdev/openui
Original link: https://x.com/RoundtableSpace/status/2116994940244803862