ag2

AG2 (formerly AutoGen) is a Python framework for building and orchestrating multi-agent AI systems, where multiple conversable agents collaborate to complete tasks. It provides core primitives including `AssistantAgent` for LLM-driven reasoning, `UserProxyAgent` for executing code and representing human input, and a suite of multi-agent conversation patterns such as group chats, hierarchical orchestration, and human-in-the-loop workflows. Claude can be configured as the backing LLM for any agent through the `LLMConfig` setup, and the framework includes native MCP support so agents can call external tools exposed via MCP servers. A notable feature is the `code_execution_config`, which lets proxy agents run generated code locally or inside Docker containers without manual intervention. The framework targets researchers building agentic AI systems, software developers automating complex multi-step workflows, and teams who need flexible agent-to-agent communication patterns across heterogeneous LLM providers. AG2 is available on PyPI and requires Python 3.10 or higher.

What people ask about ag2