AutoGen
by Microsoft
Microsoft framework for multi-agent conversational AI applications
Last reviewed 2026-06-18
AutoGen is an open-source programming framework from Microsoft Research for building agentic AI applications, centered on multi-agent conversation: customizable, conversable agents that integrate LLMs, tools, and humans and coordinate through automated agent-to-agent chat. Common patterns include two-agent chats and group chats with a manager that orchestrates which agent speaks. AutoGen v0.4 was a from-the-ground-up rewrite to an asynchronous, event-driven architecture, layered as a low-level Core API, a higher-level AgentChat API for rapid multi-agent prototyping, and extensions for model and tool integrations. As a framework, AutoGen provides the orchestration scaffolding, not a finished product: the autonomy and reliability of any system you build are determined by your implementation, your tools, and whether you include human-in-the-loop steps. It is model-agnostic (with OpenAI and other model extensions) and supports human participation as a first-class agent. The project is now community-managed and described as in maintenance mode; the original team's active line of work continues under the renamed AG2 project.
What it can do
Orchestrate multi-agent conversations
SupervisedDefine conversable agents that integrate LLMs, tools, and humans and coordinate via automated agent-to-agent chat; the resulting autonomy is developer-defined.
sourceBuild two-agent and group-chat patterns
SupervisedAgentChat provides opinionated primitives for common patterns such as two-agent chat and manager-orchestrated group chats for rapid prototyping.
sourceAdd human-in-the-loop participation
SupervisedHumans can participate as a first-class agent in a conversation, approving or steering the flow during execution.
sourceIntegrate tools and models (event-driven core)
SupervisedThe v0.4 asynchronous, event-driven Core API plus extensions integrate tools and multiple model providers, with observability and flexible control.
source
Strengths
- +Strong, well-known abstraction for multi-agent conversation (two-agent and group-chat patterns) from Microsoft Research
- +v0.4 rewrite brings an asynchronous, event-driven architecture with better observability and control
- +Open source, model-agnostic, and supports humans as first-class participants in agent conversations
Limitations
- −Framework, not a product: autonomy and reliability depend entirely on what the developer builds
- −Now community-managed and described as in maintenance mode, with the original team's active work continuing under the renamed AG2 project
- −Breaking changes between v0.2 and v0.4 mean migration effort and fragmented online examples
Overview
AutoGen is an open-source framework from Microsoft Research for building agentic AI applications, with multi-agent conversation as its signature abstraction. It is developer infrastructure, not an end-user product.
What it does
Developers define conversable agents that integrate LLMs, tools, and humans and coordinate through automated agent-to-agent chat. AgentChat offers opinionated primitives for common patterns (two-agent chat, manager-orchestrated group chats), built on a low-level, event-driven Core API introduced in the v0.4 rewrite. Humans can join as a first-class agent. The autonomy of any system is determined by the developer's design, the tools granted, and whether human-in-the-loop steps are included.
Integrations & setup
Installed via pip (autogen-agentchat plus extensions). Model-agnostic with OpenAI, Azure OpenAI, Anthropic, and local model support; tools are integrated through function calling and extensions. Runs anywhere the developer hosts it.
Pricing
The framework is free and open source. Costs are the underlying model and infrastructure the developer brings.
Project status
The microsoft/autogen repository is now community-managed and described as in maintenance mode (bug fixes, security, docs). The original team's active development continues under the renamed AG2 project, which forked from AutoGen.
Best for / not for
Best for developers who want a well-known multi-agent conversation abstraction and an event-driven core. Less suited to non-developers or anyone needing an out-of-the-box agent, and buyers should weigh the maintenance-mode status.
Alternatives
LangGraph offers graph-based stateful agents; CrewAI offers role-based multi-agent crews; LlamaIndex centers on data and RAG-grounded agents.
What people are saying
We aggregate real LinkedIn discussion into sentiment for the agents people search most. AutoGen isn't tracked yet, want it added? Request tracking.
FAQ
Is AutoGen an autonomous agent?+
AutoGen is a framework, not an agent. It provides the scaffolding for multi-agent conversations; how autonomous a system is depends on what the developer builds, which tools they grant, and whether they include human-in-the-loop steps. It supports humans as a first-class participant.
What changed in AutoGen v0.4, and is the project still active?+
v0.4 was a ground-up rewrite to an asynchronous, event-driven architecture (Core API plus a higher-level AgentChat API) with better observability. The microsoft/autogen project is now community-managed and described as in maintenance mode; the original team's active development continues under the renamed AG2 project.
Sources
- microsoft/autogen on GitHub · accessed 2026-06-18
- AutoGen documentation (stable) · accessed 2026-06-18
- AutoGen multi-agent conversation framework (0.2 docs) · accessed 2026-06-18
Last reviewed 2026-06-18