Tool not found.
Back to Home
Open-source platform for computer-use AI agents, offering background desktop drivers, cross-OS agent sandboxes, and benchmarks for training and evaluating agents that see screens and control apps.
Cua is a suite of tools for building and scaling computer-use agents across macOS, Windows, and Linux. At its core, Cua Drivers let agents drive native desktop apps in the background — clicking, typing, and verifying UI state without stealing the cursor or window focus — via a shared CLI and MCP server that plugs into Claude Code, Cursor, Codex, and other clients, with Linux support for both X11 and Wayland. The main Cua SDK provides a single Python API for spinning up agent-ready sandboxes (Linux, macOS, Windows, or Android, in the cloud or locally via QEMU) with unified screenshot, mouse, keyboard, shell, and multi-touch gesture controls. Cua-Bench complements this with benchmarks and RL environments (OSWorld, ScreenSpot, Windows Arena, and custom tasks) for evaluating agents and exporting trajectories for training. Rounding out the stack, Lume handles macOS/Linux virtualization on Apple Silicon using Apple's Virtualization.Framework, with unattended VM presets for quick setup. The project is released under the MIT License.
