AutoGen is Microsoft Research's open-source framework for multi-agent conversational AI. Its core primitive is a conversable agent — an agent that can exchange messages with other agents and optionally humans. Group chat patterns let multiple agents collaborate on tasks with configurable speaker selection, turn-taking, and termination conditions.
You define agents as subclasses of ConversableAgent, each with an LLM config and a set of tools. GroupChat aggregates multiple agents into a conversation with a manager that picks the next speaker based on prior messages. AutoGen supports human-in-the-loop patterns where a human agent can interject at configurable points. Code execution is first-class — agents can write and execute code in sandboxed environments. AutoGen Studio provides a no-code UI for prototyping multi-agent systems.
Completely free and open source. Self-host your own LLM or use any API (OpenAI, Anthropic, local). No paid tier.
Microsoft Research, academic teams, and builders doing multi-agent experiments. Less common in production than LangChain/LlamaIndex but influential in research.