CopilotKit open-sources ChatGPT's dynamic UI

Open Intelligent UI is an open-source framework that brings ChatGPT-style dynamic interfaces straight into your own web applications.

OpenAI rolled out Intelligent UI this week, allowing ChatGPT to generate interactive charts, maps, and functional widgets on demand. But that feature is locked inside ChatGPT with no public API in sight. CopilotKit built Open Intelligent UI to give developers that exact capability while honoring their own design systems and component libraries.

How it actually works

Under the hood, the system splits generative UI across a spectrum between total control and full flexibility.

For simple outputs like data tables or forms, it follows specifications like Google's A2UI, mapping responses straight to your existing frontend components. That keeps your fonts, brand colors, and established styling intact.

When a task calls for something custom, such as a mini-game or interactive calculator, it triggers generateSandboxedUi. The agent streams raw HTML, CSS, and JavaScript into an isolated iframe sandbox, rendering the UI live as the code arrives.

These sandboxed widgets are fully functional, not static mockups. Every button click or slider drag inside the widget loops back as a new conversation turn to the agent. You can also download any rendered interface as a standalone HTML file.

Open protocols and easy setup

The project runs on CopilotKit's Open Generative UI stack, with LangChain Deep Agents driving the logic. OpenAI handles content generation, while TypeSafe's Jev determines whether an answer should be raw text, a native component, or a sandboxed mini-app.

Because it adheres to the open AG-UI protocol and supports MCP Apps, you are not married to any specific stack. You can swap out the LangChain agent for Mastra, Google ADK, or Microsoft Agent Framework without touching the frontend layer.

The codebase is live on GitHub under CopilotKit/OpenIntelligentUI. You can spin it up locally in minutes using make setup followed by make dev. You just need an OpenAI API key and a TypeSafe key, both of which can be entered right inside the chat header.

Why it matters

We think this release solves the real bottleneck of generative UI: adoption within existing products.

As OpenAI designer Andrew Chen pointed out, the hardest part of generative UI isn't creating the UI itself, but making it feel like it belongs in your app. Open Intelligent UI hands teams the architectural blueprint to bridge that gap without waiting for proprietary vendor APIs.

If you are building agentic software, it is definitely worth cloning today to see how your component catalog holds up. Just keep in mind that sandbox stability still relies heavily on the code quality of your base model, so keeping production apps tied closer to controlled native components remains the safer bet for now.