cua: Scale computer-use 2.0 with open-source drivers, cross-OS sandboxes, and benchmarks for training, evaluation, and data generation.
What problem it solves
Computer-use projects need environments where agents can operate applications, run code, and interact with graphical interfaces within the same task. They also need repeatable ways to create tasks, evaluate agents, export trajectories, and generate data for training and research.
Key features
Shared desktop Spaces — Cua Spaces gives agents full desktops hosted on computers owned by the user, including macOS and Linux environments. Spaces support signed-in application teleportation, multiplayer collaboration, and agent-ready images.
Source: README
Cross-platform desktop automation — Cua Driver provides tools for inspecting and operating native applications and browsers on macOS, Windows, and Linux. Agents can connect through the CLI, MCP, or typed SDKs.
Source: README, libs/cua-driver/README.md
SDK and CLI sandboxes — The SDK and `cua` command manage local VMs and containers, including computers running cua-spacesd. The same API is available for Python, TypeScript, and Rust, with support for embedded use or a shared daemon.
Source: README, libs/cua/README.md
Local virtual machines — Lume creates and manages local macOS and Linux virtual machines on Apple Silicon using Apple's Virtualization.Framework.
Source: README
Task evaluation and specialized models — Cua Bench supports task creation, agent evaluation, and trajectory export, while CUA-S1 provides source code for small models specialized in bounded computer-use decisions. The research components include synthetic-data generation, training, evaluation, and documented deployment guidance.
Source: README
Quick start
Copied verbatim from the project's README.
curl -fsSL https://cua.ai/install.sh | shcua sb create ubuntu --name dev # a gVisor container, set up on first use
cua sb exec dev uname -a
cua sb screenshot dev
cua sb rm devfrom cua_sandbox import Image, Sandbox
async with Sandbox.ephemeral(Image.linux(), local=True) as sb:
print((await sb.shell.run("uname -a")).stdout)Who it may suit
This project may suit developers building computer-use agents, teams operating agents alongside people, and researchers creating evaluation tasks or training data.
Topics to discuss
Which desktop or sandbox workflow should receive the next integration example?
How should benchmark tasks balance simulated environments and real applications?
Which platforms and agent frameworks need clearer setup guidance?
Written by ReadmeX AI from the project's README, source and public GitHub data; it may contain mistakes.
Read the original README https://github.com/trycua/cua#readme
ReadmeX opened this community for a trending project; it is not run by the project. Maintainers can claim it from the community page.
Comments
…