The Desktop AgentOS for Windows Automation
Open-source (MIT) Desktop AgentOS to automate multi-app Windows workflows. Uses hybrid GUI+API control, RAG, & multi-agent framework for robust automation.
UFO² is positioned as stable, battle-tested, and easy to set up with a low learning curve. It serves as an ideal entry point for understanding agent architecture, ReAct loops, and practical automation before scaling to complex multi-device scenarios.
UFO² uses speculative multi-action batch predictions to reduce LLM calls by 51% and hybrid visual+UIA detection for robust control identification. This enables efficient execution of multi-step workflows on Windows with optimized token usage.
UFO² provides deep Windows OS integration (UIA, Win32, WinCOM) and hybrid GUI+API actions to automate applications and system tasks. Users can define tasks and let the agent execute them autonomously on Windows, reducing manual effort.
UFO² includes a Knowledge Substrate with RAG (Retrieval-Augmented Generation) capabilities that use docs, demos, and execution traces to enhance decision-making and reduce errors during task execution.
UFO² can serve as a Galaxy device agent within UFO³'s multi-device orchestration framework. It receives tasks from ConstellationAgent, executes Windows-specific operations, and reports results back, enabling seamless participation in cross-device workflows.