AG2 logo

AI AGENTS

AG2

by AG2

agentsource:ag2.aimulti-agentagent-frameworkopen-sourceagentosllm-toolshuman-in-the-looppython

Overview

Open-source framework for building AI agents and enabling multi-agent collaboration with LLMs, tools, human-in-the-loop workflows, and conversation patterns.

Details

AG2 is described by its official site and docs as an open-source multi-agent framework and AgentOS for building, orchestrating, and coordinating teams of AI agents with tools, workflows, and protocols. The quickstart describes it as a framework for building AI agents and enabling multi-agent collaboration with LLMs, tools, human-in-the-loop workflows, and conversation patterns. AG2 documentation also covers tools that let agents interact with external systems, fetch real-time data, execute complex tasks, and use a two-step selection-and-execution workflow.

When to Use

Build applications where multiple AI agents need to collaborate instead of relying on one assistant. Coordinate LLM-backed agents that can use tools conversation patterns and human-in-the-loop workflows. Prototype agent workflows using AG2 reference agents such as DocAgent ReasoningAgent WebSurferAgent DeepResearchAgent and WikipediaAgent.

Getting Started

  1. Read the AG2 quickstart guide: https://docs.ag2.ai/latest/docs/home/quickstart/
  2. Inspect or install the published Python package from PyPI: https://pypi.org/project/ag2/
  3. Review the main open-source repository: https://github.com/ag2ai/ag2
  4. Explore sample applications and showcases: https://github.com/ag2ai/build-with-ag2

Key Features

  • Multi-agent framework for building AI agents and enabling collaboration between them.
  • AgentOS for building
  • orchestrating
  • and coordinating teams of AI agents.
  • Support for LLMs
  • tools
  • human-in-the-loop workflows
  • and conversation patterns.
  • Tooling model for external-system interaction
  • real-time data fetching
  • complex task execution
  • and two-step tool selection/execution.
  • Reference agents including CaptainAgent
  • communication agents
  • DocAgent
  • ReasoningAgent
  • WebSurferAgent
  • DeepResearchAgent
  • and WikipediaAgent.

Capabilities

  • multi-agent collaboration
  • agent orchestration
  • tool-using agents
  • human-in-the-loop workflows
  • document and web-page RAG via DocAgent
  • web-surfing and research-oriented agent workflows

Last updated Jun 10, 2026