iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Cua is an open-source computer-use stack for building and operating AI agents that interact with desktop environments. It is not one interchangeable tool: Driver controls windows in native desktop apps, Sandbox provisions isolated environments, Fleets provides hosted capacity, Spaces is an interactive desktop app for working alongside agents, and Bench supports creating and evaluating computer-use tasks. Choose the component based on whether you need to control an existing desktop, run an isolated machine, scale hosted workloads, collaborate at a desktop, or measure agent performance.
What Cua is—and which component fits your job
Cua describes its SDK as a way to create and drive computers from code. Its documentation and repository group several related tools under the Cua name; the components have distinct roles rather than being alternative names for the same product. Cua documentation and the Cua repository are the starting points for the broader project.
| Component | What it is for | Best fit when you need to… |
|---|---|---|
| Cua Driver | Send input to and inspect windows in native desktop applications. | Have an agent operate an existing desktop app through MCP or CLI. |
| Sandbox | Provision isolated computer environments, with local backends including Docker, QEMU, and Apple Virtualization. | Give an agent a separate environment rather than control the user’s existing desktop. |
| Fleets | Hosted, usage-priced infrastructure for workloads requiring more concurrency. | Run hosted workloads instead of supplying all compute locally. |
| Spaces | An interactive desktop app for working alongside agents. | Work with an agent at a desktop environment. |
| Cua Bench | Tools for authoring computer-use tasks and running or scoring benchmarks. | Create evaluations and inspect agent performance on computer-use tasks. |
The repository also includes Lume, a local macOS VM tool, and CUA-S1 decision models. These are additional project components, not synonyms for Driver, Sandbox, Fleets, Spaces, or Bench.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
What Cua Driver does—and what it does not guarantee
Cua calls Driver an open-source computer-use driver for macOS, Windows, and Linux. It exposes interaction through an MCP stdio server, a long-running daemon, or a one-shot CLI. The product page describes connecting agents through MCP or CLI and directing commands at a chosen window. See the Cua Driver product page and its quickstart for setup and current platform details.
#1 Best Overall
Driver is designed to send input to a target window without moving the system cursor. That does not mean every app can be controlled invisibly in the background: Cua notes that some software requires focus. The driver reports this constraint, allowing a caller to retry or permit foreground input. Whether a particular action works depends on the operating system, application, and configuration.
Documented platform labels
Cua’s Driver page currently lists macOS Sequoia and Tahoe, Windows 11 and Server 2025, and Linux X11 plus XWayland. These are the support labels stated in Cua’s documentation, not a guarantee that every app or setup behaves identically. Check the live product page and quickstart for changes before installing.
Licensing scope
Cua’s Driver product page identifies Driver as MIT licensed. Do not assume that label automatically covers every part of the Cua project: the repository distinguishes components, and the optional perception extension has separate licensing notices. Check the license for the specific component and release you plan to use.
When to choose a local macOS sandbox
The local macOS VM path is for readers who specifically need a macOS environment running on their own Apple silicon Mac. Cua’s local macOS sandbox guide lists Apple silicon, macOS 14 or later, about 40 GB of free disk space, and 8 GB of free memory as requirements. Those requirements apply to this tutorial path—not to Driver generally, Linux or Windows environments, or hosted Fleets.
The guide’s example uses the Cua CLI to create a Tahoe sandbox with the Lume runtime. It describes an initial 23 GB image pull taking about eight minutes, then later clones becoming ready in about 25 seconds. These are figures reported by Cua for the documented workflow, not independent benchmarks or guaranteed timings; actual results can vary with hardware, storage, and network conditions.
Typical workflow in the Cua guide
- Install the Cua CLI using the guide’s current instructions.
- Check for or install the Lume runtime.
- Create the macOS sandbox with the CLI.
- Run a command in the sandbox, then capture a screenshot or open the desktop viewer.
- Remove the sandbox when you are finished with it.
The tutorial says its Tahoe image includes cua-spacesd and pre-granted screen and accessibility permissions. Its Sequoia example differs in command execution and viewing, so do not assume the Tahoe setup details apply to every image.
When hosted Fleets make more sense
Local Sandbox backends and hosted Fleets solve different infrastructure problems. Cua presents Docker, QEMU, and Apple Virtualization as local Sandbox options; Fleets is its hosted option for workloads that need more concurrency. The hosted service is usage-priced, so compare the current rates with the cost and operational burden of running your own compute before choosing.
As posted on Cua’s homepage on October 7, 2026, Fleet rates were $0.044625 per vCPU-hour and $0.0223125 per GiB-hour. These are Cua’s published rates on that date, not a guaranteed quote; check Cua’s homepage for current pricing and terms.
Best Value
Where Spaces and Cua Bench fit
Spaces is the desktop app in Cua’s component lineup for working alongside agents. It is distinct from Driver, which provides window-level input and inspection, and from Sandbox, which provisions an environment. Cua’s documentation also presents “Host Spaces on your spare Mac” as a task, but that phrasing alone does not establish a required hardware model or a general recommendation to buy one.
Cua Bench is the evaluation and task-authoring part of the stack: it is for creating computer-use tasks and running or scoring benchmarks. It matters when you need to assess an agent’s performance or build repeatable evaluations, rather than simply provision a desktop or connect an agent to an app.
How to decide what you need
- Control a native app already running on a desktop: start with Driver; check its input and focus behavior against your target application.
- Give an agent an isolated environment on your own machine: use Sandbox and select a backend that fits your host OS and deployment.
- Run more concurrent hosted workloads: evaluate Fleets and verify current usage pricing.
- Work interactively alongside an agent: look at Spaces.
- Author tasks or score agent performance: use Cua Bench.
For any choice, check the operating system and application support, whether foreground focus is acceptable, the integration surface you need (MCP, CLI, or SDK), where compute will run, expected concurrency, and whether you need evaluation tooling. These distinctions are reflected across Cua’s documentation, its Driver page, and its homepage.
Recommended Free Tools
Do you need to buy a Mac?
Only if you choose the documented local macOS sandbox route and do not already have suitable Apple silicon hardware. The cited guide specifies Apple silicon and other system requirements, but it does not recommend a Mac mini or name a preferred Mac model. Driver use, Linux or Windows sandboxes, and hosted Fleets do not by themselves imply that you need to purchase a Mac.
Project background
Cua says it began in San Francisco in March 2025 and describes its aim as building an open-source, self-hostable layer for computer-use agents. That is company-provided background and positioning, as stated on its About Cua page.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
