autogen: Build Multi-Agent AI Applications

Overview

AutoGen is a framework for creating multi-agent AI applications, where agents can act autonomously or work with people. Its layered APIs support both event-driven agent systems and higher-level chat and orchestration patterns, helping developers build workflows that combine language models, tools, and specialized agents.

It is most relevant to developers maintaining existing AutoGen projects or prototyping agent workflows. The project is in maintenance mode, so Microsoft recommends its successor, Microsoft Agent Framework, for new projects.

Key Features

• Core API for message passing, event-driven agents, and local or distributed runtimes.
• AgentChat API for higher-level agent interactions and common multi-agent patterns.
• Extensions API for model clients and capabilities such as code execution.
• Integrations for model providers, including OpenAI and Azure OpenAI.
• MCP workbench support for connecting agents to MCP servers and tools.
• AutoGen Studio, a no-code GUI for prototyping multi-agent workflows.
• AutoGen Bench, a suite for evaluating agent performance.
• Python and .NET resources within the project ecosystem.

Use Cases

• Developers can prototype a workflow in which a lead assistant delegates questions to specialized agents.
• Teams can connect an agent to trusted MCP tools, such as a browser server, to perform tool-assisted tasks.
• Existing AutoGen users can maintain or migrate applications built with its Core, AgentChat, and Extensions APIs.
• Researchers and developers can explore multi-agent patterns or use AutoGen Bench to evaluate agent performance.
• Teams can use AutoGen Studio to demonstrate workflows before building a production application.

Getting Started

Requires Python 3.10 or later. Install AgentChat and the OpenAI extension:

pip install -U "autogen-agentchat" "autogen-ext[openai]"

See the repository README and documentation for setup and examples.

Alternatives

• mcp-agent: Unlike AutoGen’s general multi-agent framework, mcp-agent centers on MCP integrations, composable workflows, and durable execution.
• griptape: Griptape builds agents alongside task, pipeline, and workflow abstractions, rather than centering on multi-agent orchestration.
• fast-agent: fast-agent emphasizes declarative multimodal agents and MCP workflows, while AutoGen focuses on multi-agent orchestration.
• openai-agents-js: This TypeScript SDK supports tool-using and delegating agents, while AutoGen is a Python framework with broader multi-agent orchestration.

Considerations

• AutoGen is in maintenance mode and will not receive new features or enhancements. Contributions are limited to bug fixes, security patches, and documentation improvements.
• Microsoft recommends Microsoft Agent Framework for new projects and provides a migration guide.
• AutoGen Studio is intended for prototyping and demonstrations, not production deployment. Applications require their own authentication, security, and other deployment safeguards.
• The quickstart examples use OpenAI API credentials. Only connect to trusted MCP servers, which may execute commands in the local environment or expose sensitive information.
• The README states that documentation and other content are licensed under CC BY 4.0, while code is licensed under MIT. Check the repository license files for the terms applicable to each component.