Microsoft AutoGen

Microsoft AutoGen is an open-source programming framework developed by Microsoft Research that enables developers to build sophisticated LLM-powered applications using a multi-agent conversation paradigm. Rather than treating a large language model as a single monolithic assistant, AutoGen lets you define multiple specialized agents — each with its own role, system prompt, tools, and capabilities — and have them collaborate through structured conversations to accomplish complex tasks. This approach mirrors how human teams operate, where specialists with distinct expertise coordinate to solve problems that no single member could tackle alone.

The framework originated at Microsoft Research as part of a broader effort to simplify the orchestration, optimization, and automation of LLM workflows. At its core, AutoGen provides customizable and conversable agents that can integrate LLMs, human inputs, and external tools in flexible combinations. Developers can construct simple two-agent chats (for example, an AssistantAgent paired with a UserProxyAgent that executes code) or elaborate group chats where a manager agent routes messages among a team of specialists such as planners, coders, critics, and reviewers. Agents can write and execute Python code, call functions, browse the web, query databases, and hand off work to humans when needed.

AutoGen supports diverse conversation patterns, including fully autonomous agent-to-agent dialogue, human-in-the-loop workflows where a person can intervene or approve steps, and hierarchical structures where one agent supervises others. The framework is model-agnostic, working with OpenAI models, Azure OpenAI, local open-source models via Ollama or LM Studio, and other providers through a unified client interface. It also includes built-in support for code execution in Docker containers or local environments, retrieval-augmented generation, and integration with external APIs.

The project has evolved significantly since its initial release. The modern AutoGen (v0.4 and beyond) introduces a layered architecture with AutoGen Core for event-driven agent runtimes, AutoGen AgentChat for high-level conversation patterns, and AutoGen Extensions for integrations. Alongside the Python library, Microsoft released AutoGen Studio, a low-code interface that lets users prototype multi-agent workflows visually without writing code. AutoGen has become one of the most widely adopted agentic frameworks in the open-source ecosystem, with tens of thousands of GitHub stars and an active research community publishing papers on topics like automated agent design, cost optimization, and evaluation benchmarks such as GAIA.