AG2 is an open-source framework for building AI agents and multi-agent systems that work together to solve tasks.
The framework addresses the challenge of coordinating multiple AI agents to accomplish complex objectives. It provides a protocol-driven architecture that enables agents to interact with each other, integrate with various large language models, and support tool use. The approach allows developers to construct autonomous workflows, human-in-the-loop systems, and multi-agent conversation patterns without being locked into a specific agent implementation or communication protocol.
Developers should choose AG2 if they need to build systems where multiple agents collaborate on tasks, require flexibility in agent design and LLM selection, or want to incorporate human oversight into agentic workflows. The framework suits projects ranging from research prototypes to production applications. Note that AG2 v1.0 introduced a protocol-driven framework as the primary interface, while the earlier class-based agent system (ConversableAgent, GroupChat) moved to a separate classic repository and remains maintained for existing projects.
The project is maintained by a dynamic group of volunteers from multiple organizations, with support available through designated administrators. Development activity shows active maintenance of the core framework alongside continued support for the legacy classic implementation in a separate repository.
