What it does

AutoGen UI is a minimal chat interface for Microsoft’s AutoGen multi-agent framework. A FastAPI backend exposes a single /generate endpoint that runs a predefined team of LLM agents against your prompt, then streams the results to a Next.js frontend. The whole thing is deliberately simple: one hardcoded agent team, no auth, no history database, no team switching from the UI.

The interesting bit

The project wears its limitations openly — the README literally says “this app is clearly just a starting point” and points you to AutoGen Studio for anything serious. That honesty is refreshing. What’s actually useful is seeing how the AgentChat API’s streaming fits into a modern web stack, with the manager class (autogenui.manager) handling the agent orchestration and the FastAPI layer dealing with SSE-style streaming to the client.

Key highlights

• Built against AutoGen 0.4x’s new AgentChat API (the README warns of future breaking changes)
• Agent team configuration lives in a JSON file (notebooks/default_team.json) — editable, but not from the UI
• Includes a tutorial notebook showing how to load custom team specs programmatically
• One-command install via pip install autogenui, though you still need to wrangle OPENAI_API_KEY and frontend build steps
• The author is literally the lead on AutoGen Studio, so this is effectively a stripped-down reference from the core team

Caveats

• No authentication, no authorization, no persistence — the README lists these as “ideas on how to extend it”
• Frontend requires manual rebuilds when you change backend hostname/port or modify frontend code; the CLI serves a precompiled bundle
• Python 3.9+ required; no mention of other LLM providers beyond OpenAI

Verdict

Grab this if you’re building a custom AutoGen interface and want to see how the streaming plumbing works in practice. Skip it if you need a ready-made tool — the README itself sends you to AutoGen Studio for that.