UFO³
An open-source AI agent framework under MIT for coordinating Windows, Linux and Android devices, with OpenAI and Azure OpenAI model support.

UFO³ is a self-hosted AI agent framework for developers and researchers automating tasks across Windows, Linux and Android devices. Its Galaxy coordinator splits a request into connected tasks, assigns them to suitable devices and tracks work that depends on results from another machine. The Python project is open source under the MIT license.
Galaxy can run independent tasks in parallel and revise the workflow as results arrive. It matches work to device capabilities and available resources, can reassign tasks, and includes automatic error recovery and progress tracking. This makes it relevant for workflows that span several machines or platforms, rather than stopping at one desktop.
For Windows work, Galaxy uses UFO², which can also operate as a standalone desktop agent. It combines GUI actions with native Windows API calls and identifies controls through both visual recognition and Windows UI Automation. It can draw on documents, demonstrations and previous execution traces to guide its actions.
You can extend the framework with custom device agents and MCP tools. Its Agent Interaction Protocol coordinates agents over WebSockets and supports automatic reconnection.
The agents and coordinator run on your own machines, while model reasoning uses a configured LLM API. Supported options include OpenAI and Azure OpenAI with GPT-4o, plus Qwen, Gemini and Claude. With a cloud provider, model requests leave your hardware and require access to that service.