Skip to content
Agent Guides

AutoGen / AG2 (Python)

Research-grade multi-agent framework. AutoGen pioneered GroupChat-style multi-agent conversations; AG2 is the community continuation with event-driven architecture and A2A protocol.

Install

Terminal window
pip install "ag2[openai]"

Version

AG2 · 2026 · Python 3.10+

Best for

Research, experimentation, GroupChat, human-proxy workflows.

Zero → Hero

Section titled “Zero → Hero”
1 · Introduction & overview Core ideas, mental model · 10 min
2 · Core fundamentals ConversableAgent, LLM config · 25 min
3 · Simple agents First agent, chat, code execution · 25 min
4 · Multi-agent systems GroupChat, speaker selection · 30 min
5 · Tools & function integration Registered functions, async tools · 20 min
6 · Structured output Pydantic validation · 15 min
7 · Agentic patterns Reflection, debate, planning · 20 min
8 · Memory & context Conversation memory · 15 min
9 · Context engineering Summaries, compression · 15 min
10 · AG2 2025 features Event-driven, A2A, CLI · 25 min
11 · AutoGen → AG2 migration Renames, breaking changes · 15 min
12 · Advanced multi-agent Complex topologies · 30 min
13 · Production Deployment, monitoring · 25 min

Jump to a topic

Section titled “Jump to a topic”
ConversableAgent
GroupChat
Function integration
Structured output
Agentic patterns
Memory systems
AG2 2025 features
AutoGen → AG2 migration
Observability
Advanced multi-agent

Reference

Section titled “Reference”
Comprehensive guide Single-page technical reference.
Advanced multi-agent
AutoGen → AG2 migration
Observability
Recipes
Diagrams
Production guide

What’s new

Section titled “What’s new”
  • AG2 rebrand (Nov 2024) — community continuation; legacy autogen imports preserved via aliases.
  • Event-driven architecture (autogen.beta) — typed events for complex conversations.
  • A2A protocol — cross-framework interop.
  • AG2 CLI — scaffolding and deploy helpers.

Install, minimal example

Section titled “Install, minimal example”
from autogen import ConversableAgent, LLMConfig


# Option A — inline
llm_config = LLMConfig({"api_type": "openai", "model": "gpt-4o"})


# Option B — load a classic OAI_CONFIG_LIST file or env var
# llm_config = LLMConfig.from_json(env="OAI_CONFIG_LIST")
# llm_config = LLMConfig.from_json(path="OAI_CONFIG_LIST.json")


agent = ConversableAgent(
    name="assistant",
    llm_config=llm_config,
    system_message="You are a helpful assistant.",
)


reply = agent.generate_reply(
    messages=[{"role": "user", "content": "What is 2 + 2?"}]
)
print(reply)

Ready? Start Zero → Hero Step 1 →

Revision history

Section titled “Revision history”
DateVersionChanges
2026-04-21AG2Index redesigned with Zero → Hero + Jump-to-topic grid.
November 2024AG2Rebranded from AutoGen to AG2.

Note: For the newer AG2 guide, focusing on 2025 features (event-driven, A2A, CLI), visit that framework’s page.