AutoGen
Microsoft's open-source framework for building teams of cooperating AI agents in Python or .NET; it now gets maintenance only, no new features.
What it does
AutoGen is a software library for letting several AI agents work on a task together – on their own or with a human approving steps along the way. The AgentChat layer offers ready-made patterns for group conversations between agents, while Core targets event-driven and distributed systems. Extensions connect models from OpenAI, Azure OpenAI and Azure AI Foundry, and experimentally Anthropic, Gemini and local models via Ollama. Agents can run code in Docker containers and use tools from MCP servers. AutoGen Studio is a web interface for trying things out without programming.
Who it suits
Development teams working with Python 3.10 or later or .NET who want to experiment with multi-agent workflows or keep existing AutoGen projects running. For new projects, Microsoft itself recommends the successor, Microsoft Agent Framework.
Data protection for businesses
AutoGen runs on your own machines or servers; Microsoft does not operate a cloud service for it. Where your inputs go depends on the connected model provider – server location, data processing agreement and training use are settled with that provider. With a local model via Ollama, data stays on your own hardware. There are no business plans, as the software is freely available under the MIT licence.
Limits
AutoGen now receives maintenance only, no new features. According to Microsoft, AutoGen Studio is a research prototype without sign-in or security features and is not meant for production use. Several model connections are marked as experimental.