Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/trycua/cua/llms.txt

Use this file to discover all available pages before exploring further.

Cua is an open-source Computer-Use 2.0 platform that lets AI agents operate computers — clicking, typing, reading screens, and completing tasks across native GUI applications. Whether you need to drive an existing app in the background, spin up a fresh isolated desktop, or evaluate an agent’s performance, Cua gives you the tools to do it.

Drive Your First App

Install Cua Driver and connect your agent to a real native desktop app in minutes — no focus-stealing, no cursor interference.

Start a Local Sandbox

Launch a disposable Linux desktop with Docker for fully isolated, ephemeral computer-use environments.

macOS VMs with Lume

Create and run macOS VMs on Apple Silicon with near-native performance using Apple’s Virtualization.Framework.

Benchmark Agents

Define verifiable computer-use tasks, run agents against them, and score trajectories with Cua Bench.

Choose Your Path

Cua is a suite of composable tools. Pick the one that fits your use case:

Cua Driver

Background computer-use on macOS, Windows, and Linux. Speaks MCP over stdio and doubles as a plain CLI. Works with Claude Code, Codex, Cursor, and any MCP-capable harness.

Cua Sandbox

Unified Python SDK for ephemeral and persistent OS sandboxes — Linux container, Linux VM, macOS, Windows, and Android — cloud or local.

Lume

macOS and Linux VM management on Apple Silicon. Near-native performance via Apple Virtualization.Framework with a simple CLI and HTTP API.

Cua Bench

Benchmarking and RL environments for computer-use agents. Run OSWorld, ScreenSpot, and custom tasks with trajectory export for training.

Quick Install

/bin/bash -c "$(curl -fsSL https://cua.ai/driver/install.sh)"

Understand the Platform

What Is Computer Use?

Learn how Computer-Use 2.0 differs from traditional RPA and browser automation, and how the observe-decide-act loop works.

How Sandboxes Work

Understand the isolation model, runtime backends, and lifecycle of Cua sandboxes.

How Cua Driver Works

Dive into the MCP protocol, permission model, and Rust-native runtime that powers background desktop automation.

What Is Cua Bench?

Explore the task lifecycle, evaluation protocol, and RL environment interface that Cua Bench provides.

Integrations

Connect Cua Driver to your agent harness of choice:

Claude Code

Register cua-driver as an MCP server in Claude Code.

OpenAI Codex

Attach Cua Driver to the Codex agent SDK.

Local Models

Run Ollama, llama.cpp, or MLX models with Cua Driver.

Build docs developers (and LLMs) love