Repolode
Submit
ag2ai

ag2ai/ag2

Python framework for building AI agents that use LLMs and tools, and coordinate with each other in multi-agent workflows.

View on GitHubWebsite
Stars
5,010
Forks
754
Open issues
70
Last commit
13 hours ago

What is it?

AG2 (formerly AutoGen) is an open-source Python framework for building AI agents and coordinating several agents to solve a task. An `Agent` talks to a model provider, calls Python functions decorated with `@tool`, and runs the full tool-calling loop. Provider support is installed as extras such as `ag2[openai]`, `ag2[anthropic]`, `ag2[gemini]` and `ag2[ollama]`.

For multi-agent work, AG2 v1.0 uses a `Hub` that manages agent registration, a write-ahead log and an audit trail, with agents exchanging messages over typed channels such as `conversation`, `consulting`, `discussion` and `workflow`. The project also offers a harness layer with opt-in knowledge storage, history compaction and human-in-the-loop input via `context.input(...)`. Version 1.0 is not a drop-in upgrade from the classic `autogen` namespace, which now lives in a separate repository, ag2-classic.

Key features

  • 01Agent class with async `ask()` calls and multi-turn conversation continuation.
  • 02Tool calling through plain Python functions decorated with `@tool`.
  • 03Support for multiple LLM providers installed as optional extras.
  • 04Human-in-the-loop pauses that collect input from a person during a run.
  • 05Hub-based multi-agent orchestration over typed channels with a write-ahead log.
  • 06Opt-in knowledge store and history compaction for persistent memory.

Use cases

  • Building a single assistant agent that calls custom Python tools against an LLM API.
  • Coordinating a planner and reviewer agent to refine outputs through conversation.
  • Adding approval steps where a person must confirm an action before a tool runs.
  • Prototyping agentic research or multi-agent workflows in Python.

Quick start

  1. 1

    Install the package with the OpenAI extra

    pip install ag2[openai]
  2. 2

    Set your OpenAI API key as an environment variable

    export OPENAI_API_KEY="<your-api-key>"
  3. 3

    Save the first-agent example from the README and run it

Things to know

  • Requires Python 3.10 or newer.
  • Running the examples needs a paid OpenAI API key or another model provider key.
  • Not a drop-in upgrade from AG2 Classic; the autogen import and classic agent classes were removed.

Topics

#ai-agents#multi-agent#llm#python#tool-calling#agent-framework#a2a#ag2#agentic#agentic-ai

FAQ

What is ag2?
Python framework for building AI agents that use LLMs and tools, and coordinate with each other in multi-agent workflows. AG2 (formerly AutoGen) is an open-source Python framework for building AI agents and coordinating several agents to solve a task. An `Agent` talks to a model provider, calls Python functions decorated with `@tool`, and runs the full tool-calling loop. Provider support is installed as extras such as `ag2[openai]`, `ag2[anthropic]`, `ag2[gemini]` and `ag2[ollama]`.
Is ag2 free to use? What license does it have?
ag2 is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is ag2 still maintained?
Actively maintained. The last commit to ag2 was on Oct 21, 2026.
What can I use ag2 for?
Building a single assistant agent that calls custom Python tools against an LLM API. Coordinating a planner and reviewer agent to refine outputs through conversation. Adding approval steps where a person must confirm an action before a tool runs. Prototyping agentic research or multi-agent workflows in Python.
How do I install ag2?
Install the package with the OpenAI extra: pip install ag2[openai] Set your OpenAI API key as an environment variable: export OPENAI_API_KEY="<your-api-key>" Save the first-agent example from the README and run it
Where can I get ag2?
The source code and installation instructions are on GitHub: https://github.com/ag2ai/ag2

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects