AutoGen
Free tierBuild multi-agent AI applications with conversable, customizable agents powered by LLMs
Free·Technical·Powered by OpenAI, Azure OpenAI, and others·API available·Open source
Key strengths
Multi-agent conversation orchestrationFlexible agent customization and role assignmentHuman-in-the-loop supportBroad LLM backend compatibilityOpen-source with active Microsoft Research backing
Completely free
Redmond, USA
Founded 2023
Self-hostable
No ratings yet
Developer Documentation
AutoGen provides a rich Python API for building production-grade multi-agent systems:
- Agent Classes: Core classes include
ConversableAgent,AssistantAgent, andUserProxyAgent. SubclassConversableAgentand overridegenerate_replyto implement custom agent logic. - LLM Configuration: Pass an
llm_configdict (orOAI_CONFIG_LIST) specifying model, API key, base URL, and parameters. Supports OpenAI, Azure OpenAI, Mistral, Anthropic, and local models via LiteLLM. - Tool & Function Calling: Register Python functions as tools using
@agent.register_for_llm()and@agent.register_for_execution()decorators; agents automatically invoke them during conversation. - Group Chat: Use
GroupChatandGroupChatManagerto orchestrate conversations among 3+ agents with configurable speaker-selection strategies (auto,round_robin, or custom). - Code Execution:
UserProxyAgentcan execute LLM-generated code in a local subprocess or a Docker container, with configurable safety settings. - AutoGen v0.4 Runtime: The new actor-model-based
SingleThreadedAgentRuntimeandGrpcWorkerAgentRuntimeenable distributed, event-driven multi-agent pipelines with message passing. - AutoGen Studio API: A REST API and React-based frontend for no-code agent workflow design, available as a separate
autogenstudiopackage.
