AutoGen Studio

What is AutoGen Studio?

AutoGen Studio is a low-code web app from Microsoft Research for building teams of AI agents without writing code. It runs on top of the open source AutoGen framework and gives you a browser interface to define agents, wire them into a team, and watch them work through a task together.

You describe each agent — its model, its tools, its instructions — either by filling in forms or dragging components onto a canvas, then run the team and watch the messages flow between them in real time.

What is AutoGen Studio best for?

AutoGen Studio is best for rapid prototyping — quickly sketching a multi-agent workflow, testing whether a given team structure solves a task, and debugging how agents talk to each other before you commit to writing code. It suits AI engineers, researchers, and technical builders exploring agent patterns, and non-developers who want to experiment with agentic AI hands-on. It is a prototyping and demo tool, not the runtime you ship to users.

What can AutoGen Studio do?

• Team Builder — assemble agent teams through a drag-and-drop canvas or an editable JSON specification, defining each agent’s model, tools, and prompts visually.
• Playground — run a team against a task and watch live message streaming between agents, with a visual graph of the control flow.
• Inspect the agents’ reasoning — review the “inner monologue” of each run, the artifacts produced (code, text, images), and profiling data like token counts and turn counts.
• Gallery — browse and reuse pre-built agent, model, and tool components, and share your own.
• Bring your own models — connect OpenAI, Azure OpenAI, Anthropic, and local or OpenAI-compatible endpoints through AutoGen’s model clients.
• Export to code and deploy — turn a team into a Python script, test it as an API endpoint, or containerize it with Docker for a real application.

Where does AutoGen Studio fall short?

• Explicitly not production-ready. Microsoft describes it as a research prototype “not meant to be used in a production environment” — it lacks authentication, hardened security, jailbreak protections, and access controls. For anything you ship, Microsoft points you at the underlying AutoGen framework and your own code.
• Only exposes what serializes to JSON. Because agent configurations are saved as JSON from the UI, you can only configure the serializable fields the framework exposes — anything that needs custom Python logic beyond that has to be dropped into code rather than clicked together.
• Tied to the AutoGen framework’s abstractions and pace. You build within AutoGen’s agent and team model, and the broader AutoGen framework has been in maintenance mode, so Studio moves with an ecosystem whose long-term direction is uncertain.

Is AutoGen Studio free?

Yes — AutoGen Studio is free and open source. The code is MIT-licensed and the documentation is CC-BY-4.0, with no paid tier, license key, or hosted plan. You install it with pip install -U autogenstudio and run it locally with autogenstudio ui --port 8081. Your only running costs are the model API calls the agents make (or the compute for a local model).

What does AutoGen Studio replace?

AutoGen Studio is a self-hosted, open source stand-in for managed agent-building platforms like Microsoft Copilot Studio, Amazon Bedrock Agents, and Google Vertex AI Agent Builder. Where those run your agents inside a vendor’s cloud and model catalog, AutoGen Studio keeps the agent design, orchestration, and data on your own machine and lets you plug in any model — though it targets prototyping rather than the fully managed production hosting those services provide.

FAQ

Is AutoGen Studio open source? Yes. AutoGen Studio is part of Microsoft’s AutoGen project on GitHub — the code is MIT-licensed and the documentation is under CC-BY-4.0, so you can self-host, inspect, and modify it freely.

Can I self-host AutoGen Studio for free? Yes. Install it with pip install autogenstudio and launch the web UI with autogenstudio ui. There is no paid tier; you only pay for the LLM API usage your agents generate, or nothing if you run local models.

Is AutoGen Studio ready for production? No. Microsoft states it is a research prototype for prototyping and demos, without production-grade auth, security, or deployment features. Use it to design and test agent teams, then build the production app on the AutoGen framework directly — Studio can export a team to Python code to start from.

What do I need to run AutoGen Studio? Python 3.10 or newer and pip, plus an API key for a supported model provider (OpenAI, Azure OpenAI, Anthropic) or a local/OpenAI-compatible endpoint. It installs as a single Python package and serves the UI in your browser.