Repolode
Submit
trycua

trycua/cua

Open-source computer-use toolkit giving AI agents desktops to operate, with Cua Driver, Lume VMs, cross-OS sandboxes and Cua Bench evaluation.

View on GitHubWebsite
Stars
30,573
Forks
2,216
Open issues
1,195
Last commit
3 hours ago

What is it?

Cua is a toolkit for giving AI agents computers they can operate. It covers several pieces: Cua Driver, which lets agents inspect and control native apps and browsers on macOS, Windows and Linux through a CLI, MCP or typed SDKs; Lume, which creates and manages local macOS and Linux VMs on Apple Silicon using Apple's Virtualization.Framework; and a cua SDK and CLI for running sandboxes in local VMs and containers, available in Python, TypeScript and Rust.

The repository also includes Cua Bench, a framework for building computer-use tasks, evaluating agents and exporting trajectories for training, and CUA-S1, a family of small specialized models for form-filling decisions. Cua Spaces, a desktop app for macOS, provides full desktops for agents on your own machines or cloud accounts. The core SDK, driver, Lume and bench are MIT-licensed, while Cua Spaces and Cua Driver Bench use FSL-1.1-MIT.

Key features

  • 01Cua Driver operates native desktop apps and browsers on macOS, Windows and Linux.
  • 02Lume creates and manages local macOS and Linux VMs on Apple Silicon.
  • 03The cua SDK and CLI run sandboxes in local VMs and containers.
  • 04Cua Bench builds computer-use tasks, evaluates agents and exports trajectories.
  • 05CUA-S1 provides small specialized models for computer-use decisions.

Use cases

  • Letting an AI agent drive desktop apps to automate repetitive GUI workflows.
  • Running isolated macOS or Linux sandboxes for agent experiments.
  • Benchmarking and generating training trajectories for computer-use agents.

Quick start

  1. 1

    Install the cua SDK with pip

    pip install cua
  2. 2

    Install the cua CLI and create a local Linux sandbox from the README examples

  3. 3

    Run a command inside the sandbox

    cua sb exec dev uname -a

Things to know

  • The Cua Spaces desktop app is source-available under FSL-1.1-MIT, not MIT, and Spaces is paid beyond the free individual tier.
  • Lume only runs local macOS and Linux VMs on Apple Silicon.
  • Cua Driver's background operation depends on app and platform support, with stated boundaries.
  • CUA-S1 is an early, source-only research release, and its model weights are hosted separately.

Topics

#agent#computer-use#ai-agent#desktop-automation#virtualization#sandbox#macos#apple#computer-use-agent#containerization

FAQ

What is cua?
Open-source computer-use toolkit giving AI agents desktops to operate, with Cua Driver, Lume VMs, cross-OS sandboxes and Cua Bench evaluation. Cua is a toolkit for giving AI agents computers they can operate. It covers several pieces: Cua Driver, which lets agents inspect and control native apps and browsers on macOS, Windows and Linux through a CLI, MCP or typed SDKs; Lume, which creates and manages local macOS and Linux VMs on Apple Silicon using Apple's Virtualization.Framework; and a cua SDK and CLI for running sandboxes in local VMs and containers, available in Python, TypeScript and Rust.
Is cua free to use? What license does it have?
cua is open source under the MIT license. Check the license terms before using it in commercial products.
Is cua still maintained?
Actively maintained. The last commit to cua was on Oct 22, 2026.
What can I use cua for?
Letting an AI agent drive desktop apps to automate repetitive GUI workflows. Running isolated macOS or Linux sandboxes for agent experiments. Benchmarking and generating training trajectories for computer-use agents.
How do I install cua?
Install the cua SDK with pip: pip install cua Install the cua CLI and create a local Linux sandbox from the README examples Run a command inside the sandbox: cua sb exec dev uname -a
Where can I get cua?
The source code and installation instructions are on GitHub: https://github.com/trycua/cua

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects