UFO
Open-source Windows agent that uses an LLM to operate desktop apps, and a multi-device orchestration layer cal
- Stars
- 10.2k
- Forks
- 1.1k
- Licence
- MIT
- Language
- Python
- Last update
- 6 days ago
- Type
- AI model / agent
What it is
UFO³ is a Microsoft research project for building AI agents that control Windows computers and coordinate work across several devices. Its older part, UFO², runs on a single Windows machine and uses a language model to read the screen and act on apps through clicks and API calls. The newer Galaxy part splits a request into a graph of tasks and assigns them to different devices, such as Windows, Linux, or Android, as described in the README.
- Automating multi-step tasks inside Windows applications with a natural-language request
- Splitting a request into tasks that run in parallel across several devices (Galaxy)
- Using UFO² as a Windows device agent inside a larger Galaxy setup
- Building custom device agents with the MCP-based toolkit described in the README
Great for
- Developers experimenting with LLM-driven Windows desktop automation
- Researchers studying multi-agent task orchestration
- People who want to automate repetitive tasks in Windows apps and are comfortable with setup
- Teams exploring cross-device workflows, using the Galaxy framework
Not ideal for
- Non-technical users wanting a ready-made consumer app, since the README describes setup and configuration steps
- Anyone needing a stable, fully tested production system, because the README labels Galaxy as in active development
- Users who need macOS-only automation, since the README lists Windows, Linux, and Android as supported device types
How to get it
- 1
Clone and build from source
You'll need: Python 3.10 or 3.11 · Windows for UFO² desktop automation
Clone the repository
git clone https://github.com/microsoft/UFO.git - 2
Run it
- Read the UFO² or Galaxy README in the repository for setup details
- Install the Python dependencies listed in the project documentation
- Configure your language model access as described in the docs
- Run a simple single-app task to confirm the setup works
- 3
Get the most out of it
- Start with UFO² on one Windows machine before trying the multi-device Galaxy setup
- Use the online documentation at microsoft.github.io/UFO for quick start guides
- Follow the migration guide if moving from UFO² to Galaxy
- Explore MCP integration to add tools to device agents
Downloads
Strengths
- Open source under the MIT license with a public paper and documentation site
- Two tiers: UFO² is described as simpler and stable, while Galaxy adds multi-device orchestration
- Combines GUI actions with API calls, which the README says can reduce LLM calls
Trade-offs
- Depends on a language model and its API access, which the README discusses but this summary does not detail
- Galaxy setup is described as moderate in difficulty, and it is still in active development
- Primary focus is Windows, so cross-platform reach depends on the Galaxy device options
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| UFO (this) | 10.2k | AI model / agent | advanced | 6 days ago | Developers experimenting with LLM-driven Windows desktop automation |
| self-operating-computer | 10.3k | Framework | intermediate | 17 days ago | Developers exploring AI computer-use agents |
| agents | 42.1k | Dev tool | beginner | 4 days ago | Developers who already use one of the listed AI coding tools and want ready-made specialist agents |
| RD-Agent | 15.2k | AI model / agent | advanced | 15 days ago | Researchers who want to test automated ML engineering agents on benchmark-style tasks |
| huginn | 50.3k | App | intermediate | 4 days ago | People who want IFTTT-style automations without sending their data to a third-party service |
| RemoveWindowsAI | 13.6k | Dev tool | beginner | 7 days ago | Windows 11 users who want Copilot and Recall removed |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.