AutoGen

AutoGen is Microsoft's multi-agent collaboration framework that engineers the paradigm of multiple agents conversing to complete complex tasks. It emphasizes conversational agents and composable GroupChat, and is a common baseline for both academia and early practitioners in the multi-agent orchestration space.

Core capabilities include the ConversableAgent abstraction, GroupChat orchestration, nestable tool calls, code executors, human-in-the-loop mode, and the asynchronous and event-driven architecture introduced in v0.4. It adapts to OpenAI, Anthropic, and local models.

The highlight is expressive multi-agent conversation flows, well suited for studying how agents divide and cooperate. The downside is documentation clearly lags behind code iteration, with significant API changes between 0.2 and 0.4, and old examples often need manual fixes. Behavior predictability degrades under complex topologies.

Use cases: multi-agent collaboration research, workflows requiring human-machine mixed decision-making, complex task decomposition and orchestration prototypes. Build your own observability and replay before going to production.