AutoGen Multi-Agent Framework
Build multi-agent applications that operate autonomously or collaborate with people.
- Source repo
- microsoft/autogen
- Stars
- ★ 62k
- Last updated
- 9d ago
- License
- CC-BY-4.0
- Primary language
- Python
- FA score
- 52/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A Python developer prototyping two-agent chats, group chats, or specialist delegation can assemble the workflow with AgentChat and AgentTool.
- Not a fit if
- Teams starting new projects that need new features and vendor support
- Teams needing a production-ready app with built-in auth and security
- Environments that cannot call external model APIs such as OpenAI
- Source review
- 52/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
AutoGen is a programming framework for building multi-agent AI applications that can act autonomously or work alongside humans. Its layered architecture comprises a Core API for message passing, event-driven agents, and local or distributed runtimes; an AgentChat API for higher-level orchestration; and an Extensions API for model clients, code execution, and other integrations. Python applications create AssistantAgent instances, run tasks against configured model clients, and can expose specialist agents to a coordinator through AgentTool. The ecosystem also includes AutoGen Studio for no-code prototyping, AutoGen Bench for performance evaluation, and Magentic-One as an example team that handles browsing, code execution, and files. AutoGen is now community-managed in maintenance mode, with no new features planned, and the repository directs new projects to Microsoft Agent Framework.
An application defines an AssistantAgent and supplies a model_client; the quickstart uses OpenAIChatCompletionClient to call the OpenAI API. Calling agent.run(task=...) or agent.run_stream(task=...) submits a task, allows the agent to perform configured model and tool iterations, and produces a result or streamed console messages. AgentTool wraps one AssistantAgent so that another agent can delegate work to it, as demonstrated with separate mathematics and chemistry specialists. The Core API supplies message passing, event-driven agents, and local or distributed runtimes with Python and .NET support, while AgentChat adds opinionated patterns such as two-agent and group chats. The Extensions API supplies implementations for model clients including OpenAI and AzureOpenAI, plus capabilities such as code execution. McpWorkbench can launch and connect to a Playwright MCP server so an AssistantAgent can browse the web, while AutoGen Studio provides a GUI for prototyping workflows rather than a production deployment boundary.
- A Python developer prototyping two-agent chats, group chats, or specialist delegation can assemble the workflow with AgentChat and AgentTool.
- A framework engineer who needs lower-level orchestration can use the Core API for event-driven message passing and local or distributed agent runtimes.
- A developer building a web-research assistant can connect AssistantAgent to the Playwright MCP server through McpWorkbench.
- A product or research team demonstrating a multi-agent workflow before building a custom interface can prototype it in AutoGen Studio.
- An agent researcher evaluating system performance can use the included AutoGen Bench benchmarking suite.
- A team maintaining an existing AutoGen deployment can continue applying bug fixes, security patches, and documentation updates while planning migration to Microsoft Agent Framework.
How do you install or deploy this agent?
Python 3.10 or later is required. Install AgentChat and the Extensions package with OpenAI support:
pip install -U "autogen-agentchat" "autogen-ext[openai]"Install AutoGen Studio separately if the GUI is needed:
pip install -U "autogenstudio"Before running the OpenAI examples, create an OpenAI account and export its API key:
export OPENAI_API_KEY="sk-..."The documented Playwright MCP example additionally requires its server package:
npm install -g @playwright/mcp@latestHow do you use this agent?
After setting OPENAI_API_KEY, create and run a basic AssistantAgent:
import asyncio
from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
async def main() -> None:
model_client = OpenAIChatCompletionClient(model="gpt-4.1")
agent = AssistantAgent("assistant", model_client=model_client)
print(await agent.run(task="Say 'Hello World!'"))
await model_client.close()
asyncio.run(main())To prototype and run workflows through the GUI, start AutoGen Studio:
autogenstudio ui --port 8080 --appdir ./my-appOpen http://localhost:8080 afterward. Studio is intended for rapid prototyping and interface demonstrations; a production application must add its own authentication, security, and deployment features.
What are this agent's strengths and limitations?
- The layered design offers distinct levels of control: Core for messaging and runtimes, AgentChat for accessible orchestration patterns, and Extensions for model and execution integrations.
- Python and .NET support, together with local and distributed runtimes, provides more than one implementation and deployment path.
- AgentTool turns a specialist AssistantAgent into a callable tool for another agent, giving applications a concrete delegation mechanism.
- The ecosystem includes a no-code workflow GUI, a benchmarking suite, and MCP integration in addition to the core libraries.
- The extension layer accommodates first- and third-party integrations and explicitly identifies OpenAI and AzureOpenAI model-client implementations.
- AutoGen is in maintenance mode and will not receive new features or enhancements; accepted contributions are limited to bug fixes, security patches, and documentation.
- Microsoft recommends its successor for new work, so adopting AutoGen now introduces a likely future migration requirement.
- The quickstart depends on an external OpenAI account, API key, and network service; the supplied material does not document pricing, offline model setup, or default retry behavior.
- AutoGen Studio is explicitly not production-ready, leaving authentication, security, and other deployment requirements to the application developer.
- MCP servers can execute commands locally or expose sensitive information, creating a trust boundary that adopters must manage.
- Documentation and other repository content use CC BY 4.0 while code is separately licensed under MIT, so adopters must distinguish the applicable terms.
How does this agent compare with similar options?
Microsoft Agent Framework is presented as AutoGen's enterprise-ready successor. AutoGen is community-managed in maintenance mode and receives no new features, while Microsoft Agent Framework is described as production-ready with stable APIs, long-term support, multi-provider model support, and cross-runtime interoperability through A2A and MCP. The stated direction is to choose Microsoft Agent Framework for new projects and use the migration guide for existing AutoGen applications.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| AutoGen Multi-Agent Framework This agent | 52 · Major gaps | Library / SDKFree + model costs | ★ 62k | 9d ago | Python | OpenAI API |
| aiXplain SDK | 29 · Major gaps | Library / SDKFreemium | ★ 59 | 7d ago | Python | — |
| AgentScope 2.0 | 70 · Some gaps | Library / SDKFree + model costs | ★ 35k | 2d ago | Python | OpenAI API · Claude API |
| AgenticX Multi-Agent Platform | 64 · Some gaps | CLIFree + model costs | ★ 977 | 1d ago | Python | Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: README warns that MCP servers may execute local commands or expose sensitive information, advising users to only connect to trusted servers; AutoGen Studio explicitly states it is not production-ready and requires users to implement authentication and security. However, no fine-grained permission controls, user confirmation mechanisms, or data flow transparency implementations are provided. Dependency security: CodeQL and dependency review workflows exist, but no vulnerability scan results are provided. External effects: MCP example may execute commands, but no rollback mechanism is provided. Source attribution: repository is published by Microsoft official organization, but no code signing or detailed provenance verification is provided. Deductions: lack of specific security mechanism documentation and implementation details, rollback mechanism missing.
Evidence shows: README and code examples are consistent, framework layering is clear, but no detailed error handling or failure message documentation is provided. Dependency availability: installation commands and PyPI links are provided, but no availability guarantees. Deductions: insufficient failure message and error handling documentation.
Evidence shows: README targets developers and researchers, provides multiple scenario examples (Hello World, MCP, multi-agent orchestration), and states capability boundaries (e.g., AutoGen Studio not production-ready). However, trigger precision is not detailed. Environment fit: supports Python and .NET, but no complete system requirements. Deductions: insufficient trigger precision and system requirements.
Evidence shows: README is well-structured, includes installation, quickstart, documentation links, FAQ, and known limitations (maintenance mode). License is clearly CC-BY-4.0, but code license is MIT (mentioned in README). Versioning: releases link provided, but no detailed changelog. Maintenance responsibility: clearly states community-managed and recommends migration to Microsoft Agent Framework. Deductions: insufficient changelog and version history.
Evidence shows: multiple usage examples and documentation provided, output format clear and easy to integrate. Marginal value: framework provides multi-agent orchestration and extension capabilities, but maintenance mode may reduce long-term value. Cost-benefit: open-source and free, but requires API keys and dependencies. Deductions: maintenance mode may affect long-term support.
Evidence shows: README claims are consistent with code examples, but no independent verification or test results are provided. Cross-source corroboration: documentation links and community support exist, but no third-party verification. Fact-inference separation: README distinguishes facts (e.g., maintenance mode) and recommendations (e.g., recommend MAF), but not explicitly labeled. Deductions: lack of independent verification and test evidence.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- AutoGen is in maintenance mode and will not receive new features; new projects should use Microsoft Agent Framework.
- MCP servers may execute local commands or expose sensitive information; only connect to trusted servers.
- AutoGen Studio is not production-ready; implement authentication and security before deployment.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
What's Changed * Fix docs dotnet core typo by @lach-g in https://github.com/microsoft/autogen/pull/6950 * Fix loading streaming Bedrock response with tool usage with empty argument by @pawel-dabro in https://github.com/microsoft/autogen/pull/6979 * Support linear memory in RedisMemory by @justin-cechmanek in https://github.com/microsoft/autogen/pull/6972 * Fix message ID for correlation between streaming chunks and final mes… by @smalltalkman in https://github.com/microsoft/autogen/pull/6969 …View on GitHub ↗ -
Release notes excerpt
What's Changed * Update docs for 0.7.3 by @ekzhu in https://github.com/microsoft/autogen/pull/6948 * Update readme with agent-as-tool by @ekzhu in https://github.com/microsoft/autogen/pull/6949 * Fix Redis Deserialization Error by @BenConstable9 in https://github.com/microsoft/autogen/pull/6952 * Redis Doesn't Support Streaming by @BenConstable9 in https://github.com/microsoft/autogen/pull/6954 * update version to 0.7.4 by @ekzhu in https://github.com/microsoft/autogen/pull/6955 * Update doc 0.7.4 by @ekzhu in https://github.com/microsoft/autogen/pull/6956 New Contributors …View on GitHub ↗ -
Release notes excerpt
What's Changed * Update website for 0.7.2 by @ekzhu in https://github.com/microsoft/autogen/pull/6902 * Typo in docs for 'NoOpTracerProvider' by @nicsuzor in https://github.com/microsoft/autogen/pull/6915 * Fix MCP example in readme by @ekzhu in https://github.com/microsoft/autogen/pull/6919 * Extend pydantic model capability for anyOf/oneOf item typing by @fiow123 in https://github.com/microsoft/autogen/pull/6925 * Update README.md with correct stable version by @Jp3132 in https://github.com/microsoft/autogen/pull/6942 …View on GitHub ↗