AG2 is an open-source framework for building and orchestrating AI agents, designed to streamline the development of multi-agent systems for complex task solving. It targets researchers and developers looking to create sophisticated AI applications by enabling agents to converse, utilize tools, and operate autonomously or with human oversight.

How It Works

AG2 provides a ConversableAgent as the fundamental building block, allowing agents to exchange messages and generate responses using LLMs, tools, or human input. It supports various orchestration patterns like group chats and swarms, enabling complex multi-agent workflows. The framework facilitates tool integration, code execution, and human-in-the-loop interactions, offering flexibility for diverse agentic designs.

Quick Start & Requirements

• Install via pip: pip install ag2[openai]
• Requires Python >= 3.9 and < 3.14.
• API keys for LLMs should be configured using an OAI_CONFIG_LIST file.
• Documentation: https://docs.ag2.ai/
• Examples: https://github.com/ag2ai/ag2/tree/main/notebooks

Highlighted Details

• Supports autonomous and human-in-the-loop workflows.
• Enables multi-agent conversation patterns including group chats and swarms.
• Facilitates LLM integration and tool use, including code execution.
• Offers advanced concepts like structured outputs and Retrieval Augmented Generation (RAG).

Maintenance & Community

• Evolved from AutoGen, now hosted under the AG2AI organization.
• Maintained by volunteers from several organizations.
• Discord community: https://discord.gg/pAbnFJrkgZ

Licensing & Compatibility

• Core AG2 modifications are licensed under Apache License, Version 2.0 (Apache-2.0).
• Original AutoGen code is under the MIT License.
• Apache 2.0 is generally permissive for commercial use and closed-source linking.

Limitations & Caveats

The project is actively evolving from AutoGen, with some original code retained under MIT and newer code under Apache 2.0, requiring careful attention to the LICENSE and NOTICE files for full clarity.