SuperAPI AI Directory

UFO

UFO is an open-source research project from Microsoft, comprising UFO² Desktop AgentOS for Windows desktop automation and the new UFO³ Galaxy multi-device orchestration framework. UFO² is deeply integrated with the Windows system, supports hybrid GUI and API operations, can independently complete single-device automation tasks, and can also serve as a Windows device agent for Galaxy. UFO³ Galaxy builds on UFO² by introducing cross-device collaboration capabilities, using Constellation to decompose tasks into executable DAGs, supporting dynamic graph editing, asynchronous parallel execution, and capability-matching-based device scheduling, and employing a unified AIP protocol to enable secure communication between agents. The project uses the MIT license and provides documentation, migration guides, and multi-platform device configuration instructions.

Open website

About this tool

UFO is an open-source research project by Microsoft on GitHub, positioned as a GUI agent and multi-device automation framework, overall composed of two parts: UFO² desktop AgentOS for single devices, and UFO³ Galaxy for multi-device orchestration.

In terms of purpose, UFO² is mainly used for Windows desktop automation, providing deep Windows OS integration, supporting hybrid GUI and API operations, capable of executing relatively simple tasks, and can serve as a Windows device agent in Galaxy. UFO³ Galaxy is used for cross-device collaboration workflows and complex multi-step automation, targeting scenarios that require coordinating agents across heterogeneous platforms.

In terms of main capabilities, Galaxy provides DAG-based task orchestration: Constellation decomposes requests into structured DAGs with TaskStars and dependencies, supports dynamic DAG editing to adapt to workflow evolution, and supports asynchronous execution with parallel task coordination. Its design principles include declarative decomposition into dynamic DAGs, result-driven continuous graph evolution, heterogeneous asynchronous and secure orchestration (capability-based device matching, asynchronous execution, secure locking, and formally verified correctness), a unified Agent Interaction Protocol (AIP, a WebSocket-based secure coordination layer with fault tolerance and automatic reconnection), and a template-driven device agent development toolkit integrated with MCP. Galaxy also provides capabilities such as cross-device dependency tracking, parallel execution optimization, cross-device data flow management, real-time resource monitoring and dynamic reallocation, progress tracking, and automatic error recovery. Regarding UFO², its capabilities include Windows automation, hybrid GUI and API operations, and it provides advanced features such as Automator guides, RAG configuration, and multi-action, and the documentation also mentions benchmark results on WAA and OSWorld.

In terms of usage, both frameworks require LLM API configuration. Galaxy's configuration file is located at config/galaxy/agent.yaml, and UFO²'s configuration file is located at config/ufo/agents.yaml. The model configuration guide covers more options such as Qwen, Gemini, Claude, and more. Official offerings include Galaxy quick start, Windows device setup, Linux device setup, mobile device setup (Android Agent) and device pool configuration instructions; UFO² provides installation, configuration, usage guides, and advanced feature documentation. The installation command is not currently provided in the supplied materials.

Regarding applicable scenarios, if you only need Windows automation, want to get started and learn quickly, and the tasks are relatively simple, you can start with UFO²; if you need cross-device coordination, complex multi-step tasks, and want to use advanced orchestration capabilities, you can choose Galaxy; a hybrid approach can also be adopted, using UFO² for simple tasks and Galaxy for complex tasks, and gradually migrating. UFO² has entered long-term support (LTS) status and is still actively maintained. The official statement is that there are no plans to deprecate it, and migration is gradual and optional, divided into four stages: understanding Galaxy concepts, experimenting with non-critical tasks, mixed use, and gradually migrating complex tasks. Migration is not mandatory.

Regarding limitations and status, UFO³ Galaxy is marked as under active development. The core architecture, DAG orchestration, and basic multi-device support have been completed, while advanced device types, enhanced monitoring, performance optimization, and extended documentation are still under development, making it suitable for experimental and non-critical workflows. Both UFO² and Galaxy have high extensibility, supporting custom actions and automators, custom knowledge sources (RAG), custom control detectors, custom evaluation metrics, custom device types and orchestration strategies, and visualization components. The project uses the MIT License and provides a disclaimer, security policy, and code of conduct. During deployment, attention must be paid to the configuration requirements of different devices such as Windows, Linux, and Android.