Cua: Open-Source Computer-Use Tools for AI Agents
Cua builds open-source software that lets AI agents work on real computers: seeing the screen, moving the pointer, clicking and typing in desktop apps the way a person would. Its main product, Cua Spaces, turns the computers you already own, like a Mac on your desk, a spare Linux box or machines in your own cloud account, into shared desktops that you, your team and your agents use together. Alongside it are Cua Driver for operating native apps, Lume for running macOS virtual machines on Apple silicon, an SDK and a benchmarking tool. Cua’s main GitHub repository has about 32,300 stars.
The quick answer: Cua is one of the most popular open-source options for computer-use agents, especially on macOS, and an unusual one: it doesn’t rent computers. In October 2026 it shut down its hosted Cua Cloud and moved to a bring-your-own-computer model. That makes it a strong fit for developers and teams who want agents on hardware they control, and a weaker fit for anyone who wants a managed fleet of cloud desktops. Everything below comes from Cua’s site, pricing, documentation, GitHub repository and Y Combinator profile.
Cua at a glance
Checked against Cua’s site, pricing page, developer summary, documentation and GitHub repository on November 13, 2026.
| Cua | |
|---|---|
| What it is | Open-source tools and a desktop app that let AI agents use real computers |
| Products | Cua Spaces, Cua Driver, Lume, Cua SDK, Cua Bench, CUA-S1 models |
| Where it runs | Your Mac, computers you own, or your own AWS, Google Cloud or Modal account |
| Operating systems | macOS, Linux and Windows (Windows in preview for Spaces) |
| Open source | Driver, Lume and SDK under MIT; Spaces app source-available under FSL-1.1-MIT |
| Pricing | Spaces Free; Pro $18 a month; Teams $30 per seat a month |
| Company | San Francisco; Y Combinator Spring 2025; founder and CTO Francesco Bonacci |
What Cua does
Cua’s developer summary puts it simply: “Cua builds the software that lets AI agents work on real computers. You bring the computers.” It’s open about what it doesn’t do: “Cua does not rent them.”
The pieces fit together like this:
- Cua Spaces is a desktop app that gives agents full desktops. Each desktop is a Space, a macOS virtual machine built on your Mac or a Linux image, running on your Mac, another computer you own or your own cloud account. All your Spaces appear in one list and stream to you as live desktops.
- Cua Driver lets an agent operate native apps on macOS, Windows and Linux in the background, through MCP or a CLI, often without taking over your cursor or keyboard.
- Lume creates and manages macOS and Linux virtual machines on Apple silicon using Apple’s own virtualization framework.
- Cua SDK creates and drives computers from your code in Python, TypeScript, Swift or Rust.
- Cua Bench runs and scores computer-use benchmarks on your machine, and CUA-S1 is a set of small, specialized models for computer-use decisions.
Cua describes the goal as “Computer-Use 2.0,” where an agent moves between code, APIs and graphical interfaces within the same task.
Cua Spaces for teams
Spaces is where Cua is putting its effort. Its homepage pitch is “Combine all your computers into one” and “Your computers, shared by your team and its agents.” Three features stand out:
- Teleport moves a signed-in app, such as Chrome or Slack, from your Mac into a Space so it opens there still signed in. Sessions stay encrypted on your Mac in Cua’s Keyvault, sensitive items require Touch ID, and bundles are encrypted so that, Cua says, not even its own relay can read them.
- Multiplayer lets you and your agents work on the same desktop, each with your own cursor. You step in to make a decision, then hand the desktop back.
- Agent-ready images start with permissions already granted, so an agent doesn’t get stuck at a permission dialog, and coding agents like Claude Code and Codex can use a connected model provider in every Space.
Why Cua Cloud ended
Cua previously ran a hosted sandbox service, Cua Cloud, with fleets of warm desktops. Its documentation states that “Cua Cloud and Fleets ended on 18 October 2026,” and that several older packages are no longer maintained, including one it says has known security issues and should not be run. Existing releases stay on package registries but won’t get updates.
The replacement is running Spaces locally or in your own cloud. Cua publishes extensive benchmarks from its own runs to show this is practical: a new headless Space on a Mac Studio ran its first command in about 235 to 637 milliseconds depending on the type, one Mac kept 448 idle container Spaces running, and packed Spaces on Hetzner cost at most about $0.0016 per Space-hour at list prices. Each result comes with the date, machine, method and a command to reproduce it.
Pricing
Cua’s open-source Driver, Lume and SDK are free. Cua Spaces has three plans:
| Plan | Price | Includes |
|---|---|---|
| Free | $0 | Spaces on every computer you connect, Teleport, sharing with named people, up to 2 computers in use at once and 20 remote hours a month through the hosted relay |
| Pro | $18 a month ($180 a year) | Up to 5 computers in use at once and 200 remote hours a month |
| Teams | $30 per seat a month ($300 a year) | Shared team computers, roles and device approval; each seat adds 5 computers and 200 remote hours to a shared pool |
Remote hours count only time spent using your computers from somewhere else; using a computer locally is free. Spaces in your own cloud are billed by your cloud provider, and Spaces reached on your own network don’t use Cua’s relay. Single sign-on and audit export are listed as coming next.
The company
Cua is based in San Francisco and went through Y Combinator’s Spring 2025 batch; its Y Combinator profile lists Francesco Bonacci as founder and CTO. Its main repository, created in January 2025, has about 32,300 stars. We didn’t find independent reviews beyond developer discussion of its open-source tools.
How Cua compares
Cua overlaps with computer-use platforms like Orgo and E2B, and sandbox platforms like Daytona and Tensorlake. The difference is who supplies the computers. Orgo rents persistent Linux, Windows and Mac desktops; E2B runs isolated cloud sandboxes at huge scale. Cua gives you the software and leaves the hardware to you. That means lower running costs and full control, especially for macOS, where cloud options are scarce, but more setup and no managed fleet. For agents that only need a browser, Browserbase and Browserless are simpler.
Computer-use agents are the right tool for software with no other way in. For work in the apps most people use every day, Carly connects to thousands of apps directly, so from ChatGPT, Claude or a text message it can handle email, scheduling and CRM updates without driving a desktop.
Who Cua suits
Cua is a good fit if:
- You want agents to use macOS apps, on Macs you own.
- You’d rather run agents on your own hardware or cloud account than rent desktops.
- You want open-source tools you can inspect and modify.
- Your team wants to share desktops with agents and step in when needed.
Look elsewhere if:
- You want a fully managed fleet of cloud desktops.
- You relied on Cua Cloud and need a hosted replacement.
- You want a finished assistant rather than infrastructure.
FAQ
What is Cua?
Cua is a company that makes open-source software for computer-use agents, including Cua Spaces, Cua Driver, Lume, an SDK and a benchmarking tool. It lets AI agents operate real desktops on computers you own.
Does Cua rent computers?
No. Cua Spaces is bring-your-own-compute. Cua’s hosted service, Cua Cloud, ended on October 18, 2026.
Is Cua free?
Cua Driver, Lume and the SDK are open source and free. Cua Spaces has a free plan, with Pro at $18 a month and Teams at $30 per seat a month.
What is Lume?
Lume is Cua’s open-source tool for creating and managing macOS and Linux virtual machines on Apple silicon Macs using Apple’s virtualization framework.
Can Cua run agents on Windows?
Cua Driver supports macOS, Windows and Linux. Windows support for Cua Spaces is in preview.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
