CopilotKit releases Open Generative UI repo for sandboxed charts, diagrams, and 3D widgets

• CopilotKit launched an open-source Open Generative UI template for agent-generated interfaces, positioning it as a version of Claude’s new generative UI pattern that developers can run in their own stacks.repo link
• The core output is live HTML and SVG rendered inside a “sandboxed iframe,” with examples spanning algorithm visualizations, charts, diagrams, and 3D components according to the launch post and the GitHub repo.
• CopilotKit’s repository describes a split architecture: a Next.js frontend plus a Python agent, using CopilotKit and LangGraph to decide when to answer in text, call tools, or render a widget.repo summary
• Demo clips from a chat UI demo and an interactive charts demo show the intended workflow: ask in chat, then get manipulable forms, charts, or diagrams back instead of static text.

CopilotKit’s launch post says the repo is a template for “rich, interactive AI-generated UI” built with CopilotKit and LangChain’s LangGraph.launch post The examples are concrete: “visualize algorithms,” “create 3D animations,” “render charts,” and “generate interactive diagrams,” all returned as live frontend output rather than screenshots or code snippets.launch post

The linked GitHub repo fills in the implementation details missing from the tweet. It describes a monorepo with a Next.js frontend and a Python-based agent, plus hooks for component generation, frontend tools, and human-in-the-loop flows.repo summary The same repo summary says rendering happens through a widgetRenderer path that places generated HTML into sandboxed iframes, with auto-sizing and theming support handled on the frontend.repo summary

The practical pitch is chat-native UI generation. In CopilotKit’s launch demo, a prompt to “visualize quicksort” turns into an interactive sorting animation, and the same reel ends with a rendered 3D object rather than a textual explanation.demo video A separate Generative UI demo shows the model producing a structured form with dropdowns and fields directly in the conversation.

The charting path looks equally aimed at operational and educational tools. The charts demo shows a bar chart labeled “LLM Cost by Token Type” morphing into a token-flow style diagram, suggesting the stack can swap between visualization types while staying interactive.charts demo For engineers, that makes this less about a one-off Claude clone and more about an open template for agent UIs that need custom visual outputs before platform vendors standardize the feature set.