# Open Intelligent UI

> Open-source chat interface that turns questions into explanations, comparisons, and working interactive tools using an agent and generative UI.

Open Intelligent UI is an open-source chat interface from CopilotKit, built with CopilotKit and AG-UI. The agent answers a request with plain text, a native component, or a custom interactive UI, depending on what suits the task. The project is MIT licensed, and users clone it, connect their own model and visualization router, and adapt the interface and agent to their workflows.

## What It Is

It is a generative UI chat application. Users ask a question, explore an idea, compare options, or request a small tool. Results appear in the conversation as text, native components, or generated interactive answers. Demo scenes in the README include a 3D airplane explaining pitch, roll, and yaw, comparison tables turned into charts, a bill-splitting calculator, and a coastal trip map with numbered pins.

## How It Works

A Python Deep Agent with a task-first system prompt and focused skills runs behind a FastAPI service. CopilotKit carries the agent stream to a Next.js frontend. Native components handle supported structured tasks. A `generateSandboxedUi` tool streams custom HTML, CSS, and JavaScript into an isolated iframe, where local controls run. A user-clicked follow-up can send selected values through a validated host bridge and start another agent turn.

## Model and Visualization Routing

The default answer model is `chat-latest`, which needs an OpenAI key. Jev (`jev-latest`, requiring a Jev API key) chooses the renderer for each turn. Basic tables use A2UI, while charts, diagrams, calculators, and maps use Open Generative UI. Visitors can enter their own keys in the chat header. According to the README, keys stay in browser memory until refresh or clear and are not saved to browser storage.

## Setup Path

Prerequisites are Node.js 22+, pnpm 9+, Python 3.12+, and uv. After cloning, `make setup` and `make dev` start the app on localhost:3000, with an agent health endpoint on port 8123. An optional standalone MCP server exposes skill resources, prompt templates, and an `assemble_document` tool that returns an HTML document as text for a compatible host to render.

## Features
- Chat interface that chooses text, native components, or custom interactive UI per request
- Sandboxed streaming generation of HTML, CSS, and JavaScript in an isolated iframe
- Interactive 3D explanations, tables and charts, calculators, and maps
- Bring-your-own OpenAI and Jev API keys held in browser memory
- Visualization routing between A2UI and Open Generative UI
- Optional standalone MCP server with skill resources and prompt templates
- Python Deep Agent with FastAPI backend and Next.js frontend

## Integrations
CopilotKit, AG-UI, OpenAI, Jev, A2UI, Open Generative UI, LangSmith, MCP, Docker

## Platforms
LINUX, IOS, WEB, API

## Pricing
Open Source

## Links
- Website: https://opengenerativeui.copilotkit.ai/
- Documentation: https://github.com/CopilotKit/OpenIntelligentUI/blob/main/docs/README.md
- Repository: https://github.com/CopilotKit/OpenIntelligentUI
- EveryDev.ai: https://www.everydev.ai/tools/open-intelligent-ui