Agent SwarmCompare AutoGen, Microsoft's framework for building multi-agent apps in code, with Agent Swarm, a self-hosted operating system for delegated work.
The real comparison is not which abstraction is nicer. It is whether you want to build an agent system or run a persistent agent team.
AutoGen is Microsoft's framework for building conversation-driven multi-agent applications in code. As of 2026-10-28, its README marks it as archived and no longer maintained by Microsoft, and new users are pointed to Microsoft Agent Framework. Agent Swarm is an open-source operating system for delegated work. A lead agent routes tasks to workers such as Claude Code or Codex, each in an isolated container, with shared memory, schedules, and review gates. The key difference is build versus run, and conversation versus lead-and-worker. AutoGen gives you libraries to write your own agent system. Agent Swarm is a running system you hand work to. Choose AutoGen when you are building a multi-agent application in code and want conversational agent patterns. Choose Agent Swarm when you want a standing team that takes delegated work, remembers prior sessions, and self-hosts under MIT.
AgentChat is AutoGen's high-level API for single- and multi-agent conversational applications, including two-agent chats and group chats where agents take turns in a shared conversation. It fits when the core of your product is agents talking to each other and you want to prototype that pattern quickly in Python. AutoGen Studio adds a web UI for prototyping without writing code.
AutoGen Core is the lower layer: an event-driven runtime for agents that exchange messages. It runs standalone or as a distributed multi-process runtime. Teams that want to own message routing, agent lifecycle, and deployment in code can build on it directly. The surrounding operations, such as scheduling, persistence, and monitoring, are then yours to build.
AutoGen was pioneered in Microsoft Research, and it opened up experimentation with multi-agent orchestration patterns that the wider community built on. Its design and examples are a useful reference for studying how agents debate, critique, and hand off work in dialogue. If your goal is to explore those patterns rather than run daily work, the framework was built for that.
AutoGen suits teams that standardize on Microsoft tooling. The README directs new users to Microsoft Agent Framework, which it describes as the enterprise-ready successor. A new Microsoft-stack project should evaluate Agent Framework first. Teams already running AutoGen have a published migration guide.
Agent Swarm has a lead agent that receives tasks from Slack, repositories, issue trackers, email, or the API, then delegates to workers with priority queues, dependencies, and progress reporting. You assign a goal, and the lead decides which worker takes it and in what order. That routing is built in rather than designed as message flows in your code.
Each agent keeps persistent memory and identity. The swarm also shares a searchable filesystem and a reusable skill system. Decisions and failed approaches from earlier runs stay available to later ones, so work does not start cold each time. In AutoGen, memory is a component you wire into your agents yourself.
Scheduled tasks run on cron, interval, or delayed triggers, and DAG workflows chain steps with fan-out. Recurring reports, research briefs, and maintenance runs keep going without someone starting them by hand. In AutoGen, scheduling and workflow control live in application code that you own.
Review gates and human-in-the-loop approval nodes are part of the workflow model, so a worker's output can wait for sign-off before the next step runs. This matters when delegated code changes or customer-facing text need a check before they ship. AutoGen supports human input inside agent conversations, but the approval process around it is yours to build.
Agent Swarm is MIT-licensed and runs anywhere Docker runs. You can self-host it with Docker Compose or Helm on Kubernetes, and the stack can run air-gapped. The onboarding wizard generates the compose files and checks health. Workers run Claude Code, Codex, opencode, pi-mono, Devin, or ACP-compatible agents.
A useful comparison says where each tool actually wins. agent-swarm.dev is for a persistent, owned operating team; the alternative wins when its shape fits your job better.
AutoGen is a library you embed in your own application. Agent Swarm is an operated system, not an SDK, and it does not replace framework code inside your product. If the agents are a feature you ship to your users, an embeddable framework is the right shape.
As of 2026-11-03, microsoft/autogen has 61,687 GitHub stars against 990 for Agent Swarm. AutoGen also comes from Microsoft Research, and its documentation, examples, and third-party write-ups are far more extensive. Agent Swarm is the younger project.
Agent Swarm is open source under MIT, and the fastest way to judge it is a real task. The bunx @desplega.ai/agent-swarm onboard wizard collects credentials, generates the compose files, starts the stack, and verifies health. Hand the swarm a real task from Slack or GitHub, then compare that with scaffolding the same job as an AutoGen application.
AutoGen is archived. As of 2026-10-28, its README says the project is archived and no longer maintained by Microsoft, which provides no bug fixes, security fixes, new features or support. The latest release, python-v0.7.5, was published on 2025-09-30, and the last push to the repository was on 2026-10-28. Existing users are pointed to the AutoGen to Microsoft Agent Framework migration guide.
For a new project, start with Microsoft Agent Framework. The AutoGen README directs new users there and describes it as the enterprise-ready successor, with A2A and MCP interoperability. Stay on AutoGen if you already run a working system and have no reason to migrate yet, and use the migration guide when you do. Agent Swarm is a separate choice: it runs delegated work rather than providing a framework to code against.
Only for one use case. AutoGen is a framework you embed to build multi-agent applications in code. Agent Swarm is an operating system you run: a lead agent takes delegated work, routes it to workers in isolated containers, and keeps shared memory, schedules, and review gates. If you want to build a conversational agent product, AutoGen or Agent Framework is the closer fit. If you want a standing team that does the work, Agent Swarm replaces the build-it-yourself path.
Yes. A product team can keep agent features built on AutoGen, or move them to Agent Framework, while Agent Swarm runs coding, research, and recurring operational work from Slack, GitHub, or the API. Agent Swarm's workers run Claude Code, Codex, and other harnesses, so it does not require rewriting application code that already uses AutoGen.
This page compares product categories and operating models from public product documentation and repositories. We do not claim the tools are interchangeable.
See how agent-swarm.dev stacks up against the rest.
/ Install
01 · THE EASY WAY
It reads the skill, asks about your setup, then installs the swarm.
Read https://agent-swarm.dev/skill.md and help me install and operate Agent Swarm on our infra. Ask about my setup before running anything.
02 · THE FAST WAY
One command. Needs Docker.
$ curl -fsSL https://agent-swarm.dev/install.sh | bash
Rolling it out across a team? Book a free 15-min call