UFO is a Microsoft Research project (not a shipping product) that enables AI agents to control Windows desktop applications using natural language. The framework has evolved through three versions: UFO (Feb 2024, dual-agent GUI interaction), UFO2/Desktop AgentOS (Apr 2025, centralized HostAgent with specialized AppAgents, hybrid UI Automation plus vision-based control, and Picture-in-Picture virtual desktop isolation), and UFO3 Galaxy (Nov 2025, multi-device orchestration across heterogeneous platforms via the Constellation framework). It is open source on GitHub under Microsoft Research. Key features: - Dual-agent framework (HostAgent for orchestration + AppAgents per application) - Hybrid control detection combining Windows UI Automation with vision-based parsing - Picture-in-Picture interface for automation in isolated virtual desktops (UFO2) - Multi-device orchestration across heterogeneous platforms via Constellation (UFO3) - Native API integration and domain-specific knowledge per app agent - Fully automated execution without human intervention via control interaction module
Free and open source (research project). No commercial pricing.