AG2 (AutoGen Evolved)

AG2 is an open-source multi-agent conversation framework that evolved from Microsoft's AutoGen project. After AutoGen's transition within Microsoft, the AG2 community fork emerged to continue the original vision of conversational multi-agent orchestration with a focus on community governance, enhanced modularity, and backward compatibility with existing AutoGen codebases.

The framework enables developers to build systems where multiple AI agents engage in conversations to solve complex tasks. Each agent can have different roles, capabilities, and model configurations. Agents communicate through a structured conversation protocol, with support for human-in-the-loop interactions where human participants can join agent conversations at any point.

AG2 provides several agent types including ConversableAgent (the base agent with LLM and tool use), AssistantAgent (optimized for task completion), UserProxyAgent (representing human participants), and GroupChat (orchestrating multi-agent conversations). The framework handles conversation management, turn-taking, termination conditions, and message routing automatically.

Code execution is a core capability — agents can write, execute, and iterate on code in sandboxed environments. This makes AG2 particularly powerful for coding tasks, data analysis, and any workflow that benefits from executable solutions. The framework supports Docker-based code execution for security.

AG2 has added enhanced features beyond the original AutoGen including improved agent orchestration patterns, better tool integration, enhanced memory management, and new specialized agent types. The framework integrates with multiple LLM providers including OpenAI, Anthropic, Google, and local models. The community-governed development model ensures the framework evolves based on user needs rather than corporate strategy shifts.