Overview

OpenUI is an open‑source framework that lets AI systems return interactive user interfaces instead of plain text. By running a single scaffolding command, `npx @openuidev/cli@latest create`, you obtain a starter project that already contains the rendering engine and a bundle of example components. The goal is to make the output of generative models feel like it was written by a frontend developer.

The core of the system is a compact, streaming‑first language called OpenUI Lang. Each line of the language creates one component definition. For instance, the line `root = Stack([header, chart])` produces a vertical stack containing a header component and a chart component. When the model writes a response, it can start sending the first line and the UI can begin rendering immediately, without waiting for the rest of the text. In benchmark tests OpenUI Lang used about 1,362 tokens per screen, which is 58 % fewer tokens than a comparable JSON format and 52 % fewer than the A2UI JSON standard. The LLM can output this language for any type of interface: dashboards, forms, tables, charts, or complete screen layouts. The framework can interpret the same output in multiple frontend frameworks, React, Next.js, Vue, Svelte, Angular, Nuxt, Remix, Astro, and even React‑Native, or in a plain script tag for embedding in email or other lightweight contexts.

OpenUI is designed to work with the same models you already use in your projects. It supports OpenAI, Anthropic, Gemini, DeepSeek, xAI, Mistral, Qwen, and Ollama, whether the model resides on a public cloud or on your own machine. The library can be plugged into existing SDKs like Vercel AI SDK, LangChain, LangGraph, Convex, Cloudflare, CrewAI, n8n, or Haystack through a simple instruction file generated by `npx @openuidev/cli@latest setup-agent`. This means you do not need to rewrite your agent code; just add the OpenUI instructions and the agent will be able to produce UI‑ready output. The same approach applies to coding agents such as Codex, Cursor, and Claude Code.

The OpenUI Gateway is an optional production component that accepts streaming responses from your LLM, repairs any malformed output before it reaches the user and can fall back to a different provider if the original fails.