AutoGen Framework

AutoGen is a framework by Microsoft Research for building multi-agent conversational AI systems with autonomous collaboration

ConversableAgent is the base class for all agents with send/receive capabilities and conversation management

AssistantAgent uses LLMs for generating responses, tool use, and code generation with autonomous decision-making

UserProxyAgent executes code safely and serves as the interface between humans and other agents

Four conversation patterns: two-agent chat (simple back-and-forth), sequential chat (pipeline A→B→C), group chat (manager coordinates), nested chat (sub-conversations)

Group chat uses a manager agent to dynamically decide who speaks next based on conversation context

Speaker selection strategies: auto (LLM-based), round robin (fixed order), manual (human selects), custom function (domain logic)

Set max_round (typically 10-20) to prevent infinite loops and control API costs in multi-agent conversations

Specialize agents with clear names and descriptions to help the LLM manager make better speaker selection decisions

Human-in-the-loop enables approval workflows, manual intervention, and feedback at critical decision points

Code execution runs in isolated Docker containers with automatic result extraction and iterative debugging support

Response caching reduces API calls by 40-70% and improves speed by matching semantically similar queries

Nested chat allows agents to spawn internal sub-conversations before responding to the main conversation

Termination conditions (max turns, TERMINATE message, custom logic) prevent runaway conversations and control costs

AutoGen simplifies multi-agent orchestration compared to manual coordination by providing high-level abstractions and automatic conversation management