AG2 (formerly AutoGen) is an open-source multi-agent AI framework that emerged as a community-driven fork of Microsoft AutoGen, founded by the original creators Chi Wang and Qingyun Wu after they departed Microsoft to establish independent governance. It solves the challenge of building collaborative AI agent systems by providing an AgentOS where multiple specialized agents can interact, share context, and work together to solve complex tasks. AG2 maintains backward compatibility with AutoGen 0.2 while pushing forward with new features under Apache 2.0 licensing and open community governance.
AG2 preserves the familiar conversational agent architecture from AutoGen while introducing AG2 Beta, a ground-up redesign built around streaming and event-driven architecture for production agentic systems. The framework supports multi-agent conversation patterns, tool use with any LLM provider, autonomous and human-in-the-loop workflows, group chat orchestration, and teachable agents that improve over time. AG2 provides built-in support for code execution, function calling, nested conversations, and customizable agent behaviors with a focus on stability and production readiness.
AG2 is designed for AI developers, researchers, and enterprise teams who want a community-governed, open-source alternative for building multi-agent systems without vendor lock-in to Microsoft or any single cloud provider. It integrates with Google Cloud Vertex AI Agent Builder and supports deployment across diverse infrastructure including local environments, cloud platforms, and edge devices. The AG2 community actively contributes extensions, tutorials, and integrations, making it a strong choice for teams that value open governance, transparency, and long-term independence in their agentic AI stack.