Developers - Build AI Agents with AG2 | AG2

The fastest path from idea to deployed agent.

The open-source multi-agent framework built by the creators of AutoGen. Backed by research and ready for production.

50k+ GitHub Stars
20k+ Developers

Deploy Any Agent in 60 Seconds

Intelligent, composable agents that collaborate through shared context.

pip install ag2[openai]
from autogen import ConversableAgent, LLMConfig

coding_agent = ConversableAgent("coding_agent", llm_config=LLMConfig(config_list={"model": "gpt-5-nano"}))

response = coding_agent.run(message="Write a Python function to calculate fibonacci and when you're done say 'TERMINATE'.")

response.process()

Key Capabilities

Everything You Need to Build Agents

01

Agent Orchestration

Specialized agents that work together. Define roles, share context, validate outputs.

02

Conversation Patterns

Group chat, supervisor, round-robin, auto-selection. Or define your own.

03

Human in the Loop

Observable reasoning. Intervention points. Context-aware handoffs.

Ecosystem

Built for Interoperability

AG2 provides native support and integration with leading agent protocols and standards, enabling your agents to connect with the broader ecosystem.

AGNTCY
Open standard for agent interoperability. Build agents that can communicate and collaborate across frameworks.

A2A Protocol
Agent-to-Agent communication protocol. Enable seamless message passing and coordination between agents.

Resources

Documentation

API reference, comprehensive guides, and best practices for building production-ready agents.

Examples & Templates

Ready-to-run agent patterns

Blog & Deep Dives

Technical tutorials and architecture guides

Releases

Version history and changelog

Roadmap

Upcoming features and plans

Quick Start Tutorials

Step-by-step learning paths

Research

Research-Backed, Production-Proven

Built on AutoGen's research foundation. Powering published research at leading universities.

Frequently asked questions

Build Agents with AG2

Get early access and shape the future of multi-agent systems with us.