cua
An open-source framework that gives AI agents real desktops they can see and operate (click, type, run apps) on Mac and Linux, plus local VMs, specialized small models for computer-use decisions, and a benchmark for evaluating how well agents handle real computer tasks.
View trycua/cua on GitHub- GitHub stars
- 32,283
- Forks
- —
- Language
- Rust
Project details
- Repository
- trycua/cua
- Licence
- MIT
- Primary language
- Rust
- Catalog status
- Available
- Catalog check
- 2026-11-13
Repository description
Scale computer-use 2.0 with open-source drivers, cross-OS sandboxes, and benchmarks for training, evaluation, and data generation.
Explore by topic
More tools in Automation & workflows
- agent-desktop
vercel-labs · Apache-2.0
Lets an AI agent see and click on real desktop app elements via the OS accessibility tree instead of guessing from screenshots, so it can reliably operate macOS,…
- bama
daveebbelaar · MIT
BAMA gives AI coding agents a set of skills to research, clone, design, build and test a native Mac app from scratch.
- macos-harness
browser-use · MIT
macOS Harness gives an AI agent direct, raw control of a Mac, letting it see, click, type into, and script real running apps itself instead of relying on a library of…
- nuphus-mcp
mrpulor-gh · MIT
Gives AI agents actual computer-use capability: they can see your screen and control windows, mouse, keyboard, and Chrome through a lightweight local server.