BREAKING
Windows-Use drives GUI via LLMs
How Windows-Use reads the screen
1Read UIA tree
2LLM decides
3Click / type / scroll
UIA vs vision-based agents
Windows-UseUIA tree
No computer vision
Resilient to layout shifts
use_vision=False default
Computer Usevision
Reads screenshots
Unstable at high-DPI
Font differences hurt it
0M+
Windows-MCP users
0s
min latency
0s
max latency
MIT license, Windows 7 to 11
Run it in a sandbox
AI NEWS BLITZ
An open-source AI agent now controls Windows using large language models, no computer vision needed.