Skip to content
Get packs

Guide · Open source

How to use Cua: install Cua Driver, Lume and Cua Spaces

Updated · written with AI from the project's own documentation and checked against our catalogue data

Cua is an open-source set of tools that gives AI agents a computer to work on. This guide explains what each part does, how to install the main pieces from the official commands, and how to run a first task. It is written for developers and technical users comfortable with a terminal.

Cuatrycua/cua · ★ 31.6k · MIT

Quick start

  1. 1Decide which part you need. Choose Cua Driver for agents that operate apps, Lume for local VMs on Apple Silicon, Cua Spaces for full desktops, or Cua Bench for evaluation.
  2. 2Check prerequisites. Confirm your OS. Lume needs Apple Silicon, the Cua Spaces installer needs macOS 26 or later, and Windows installers use PowerShell.
  3. 3Install the component. Run the matching install command from the README for your platform.
  4. 4Grant permissions. Allow the system permissions the quickstart asks for so the agent can see and control the app.
  5. 5Connect your agent. Use the integration guide to link Claude Code, Codex, Cursor or another agent to Cua Driver.
  6. 6Run the Calculator check. Ask the agent to compute 6 × 7 in Calculator and verify the display shows 42.

What Cua includes

Cua is a group of tools rather than a single app. Cua Spaces gives agents full desktops on your Mac and on other machines you own. Cua Driver lets agents inspect and operate native apps and browsers on macOS, Windows, and Linux. Lume creates and manages local macOS and Linux virtual machines on Apple Silicon. Cua Bench is for evaluating computer-use agents.

Pick the part that matches your goal. You do not need all of them to start.

  • Cua Spaces: full desktops for agents, on your Mac or machines you own
  • Cua Driver: agent tools for operating apps and browsers
  • Lume: local macOS and Linux VMs on Apple Silicon
  • Cua Bench: create tasks, evaluate agents, and export trajectories

Install Cua Driver on macOS, Linux or Windows

Cua Driver is the usual starting point for building agents that work in desktop apps. On macOS or Linux, run the install script from the README. On Windows, use PowerShell.

Cua Driver connects through the CLI, MCP, or typed SDKs. The README notes that background delivery lets agents work without moving your pointer or taking focus when the app and platform support it, so check platform support for the limits.

  • macOS or Linux: run the bash install command
  • Windows: run the PowerShell command in a PowerShell window
/bin/bash -c "$(curl -fsSL https://cua.ai/driver/install.sh)" irm https://cua.ai/driver/install.ps1 | iex

Run your first task with Cua Driver

The README's first example is simple and checkable. Connect your agent, ask it to compute 6 × 7 in Calculator, and have it verify that the app shows 42. The official quickstart covers platform setup, permissions, and agent connection.

Grant the permissions the tutorial asks for before you expect the agent to click or type. If your agent is Claude Code, Codex, Cursor, OpenClaw or another tool, the README links to a guide for connecting it.

  • Ask the agent to open Calculator and compute 6 × 7
  • Ask it to confirm the display reads 42
  • Follow the quickstart for permissions and agent connection

Create a local macOS or Linux VM with Lume

Lume creates and manages local macOS and Linux virtual machines on Apple Silicon, using Apple's Virtualization.Framework. You need an Apple Silicon Mac. The README's first-run example creates a vanilla macOS Tahoe VM from an Apple restore image, starts it, and connects over SSH.

The tutorial uses the Lume CLI directly and explains the unattended setup defaults. Read those before you run it, since they control how the VM is configured.

  • Requires Apple Silicon
  • Creates a macOS Tahoe VM from an Apple restore image
  • Connect to the running VM over SSH
/bin/bash -c "$(curl -fsSL https://cua.ai/lume/install.sh)"

Install the cua CLI and create a sandbox

The Cua SDK and CLI give you one command, cua, for local VMs and containers, and for any machine that runs cua-spacesd. On macOS, the Cua Spaces installer also adds the cua CLI and selects the Cua Spaces app by default. The README says the Cua Spaces installer requires macOS 26 or later.

During installation the script runs cua auth login, which signs you in and can set up AI coding agents. The Cua Driver component is off by default, so add it with the option shown in the scripts/install/README.md if you want it.

  • macOS 26 or later for the Cua Spaces installer shown in the README
  • Windows uses PowerShell for the installer
  • Cua Driver can be selected with the --select cua-driver option
curl -fsSL https://cua.ai/install.sh | sh cua sb create ubuntu --name dev

Evaluate agents with Cua Bench

Cua Bench is for evaluating computer-use agents. The README describes it as a way to create tasks, evaluate agents, and export trajectories. Its quickstart shows how to create and verify a simulated task.

Use Cua Bench when you need repeatable tests of an agent instead of one-off demos. The README does not give benchmark results, so do not treat the tool as a score source by itself.

  • Create a simulated task
  • Verify the task
  • Export trajectories for analysis

What people use it for

  • Developers building agents that operate desktop apps or browsers
  • Teams evaluating computer-use agents with benchmarks
  • Creating a local macOS or Linux VM on Apple Silicon for agent work
  • Letting an agent work in apps while you keep using your own computer

Frequently asked questions

Is Cua free and open source?

The project is listed under the MIT licence. The README says the Cua Spaces app is source-available under FSL-1.1-MIT, which is a different licence, and it describes Spaces as free for individuals with paid Pro and Teams tiers. Check the licensing section and pricing page before you rely on it.

What do I need to install Cua?

You need a terminal and a supported OS for the component you pick. Cua Spaces installer requires macOS 26 or later, Lume requires Apple Silicon, and Windows installers run in PowerShell.

Can I use Cua on Windows?

Cua Driver supports macOS, Windows and Linux, and has a PowerShell installer. The README says Cua Spaces on Windows or Linux is less clearly documented, so check the docs before planning around it.

What are the alternatives to Cua?

The listing names OpenAdapt, the Anthropic computer-use demo and OpenInterpreter as comparable projects. Cua is aimed at giving agents full desktops, native app control and evaluation tools together.

Why does my agent not control the app?

The README says Cua Driver needs permissions and agent connection to work. Check that the system permissions in the quickstart are granted and that your agent is connected through the integration guide, then retry the Calculator check.

Do I need coding skills to use Cua?

Yes, in practice. Setup involves terminals, permissions and agent configuration, so it suits developers more than non-technical users looking for a finished product.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Cua suits developers building agents that operate desktop apps, teams that want to evaluate computer-use agents, and Apple Silicon users who want local macOS or Linux VMs. It is a poor fit for simple chatbots or text-only automation, and for non-technical users who want a finished consumer app.

Install options & downloads