OpenUI is an open-source framework from Thesys that lets AI agents answer with interactive interfaces instead of walls of text. The model writes OpenUI Lang, a compact, streaming-first language in which each line declares a component, and the interface renders from your own component library while the response is still streaming. Version 1.0 of the core packages shipped on npm in October 2026 and was announced on 2026-10-15. Not to be confused with W&B's OpenUI prototype tool, which has its own aicoolies page.
OpenUI is model-agnostic (OpenAI, Anthropic, Gemini, DeepSeek, Mistral, Qwen, Ollama and others) and works with agent frameworks such as the Vercel AI SDK, LangChain and LangGraph. Renderers exist for React, Vue and Svelte, and it can use design systems like shadcn/ui, Material UI or Ant Design. npx @openuidev/cli@latest create scaffolds an app, and setup-agent adds OpenUI instructions to AGENTS.md for Codex, Cursor, Claude Code and similar coding agents.
Thesys reports that OpenUI Lang needs fewer output tokens than JSON-based UI formats (for example 58% fewer than json-render and 52% fewer than A2UI on its own 46-interface benchmark); these are vendor figures.
The framework is free and open source under the MIT license. Thesys also runs OpenUI Gateway, a managed service that repairs invalid output while it streams and fails over between providers: Free ($0, 3K API calls/month, bring your own key), Build ($49/month billed annually, 25K calls), Grow ($499/month billed annually, 500K calls, SSO/SAML) and custom Scale plans. LLM tokens are billed separately at provider rates. OpenUI Observability is free during early access.
