Cua is an open-source computer-use stack for building AI agents that interact with desktop environments. Its parts serve different jobs: Cua Driver sends input to native app windows, Sandbox provisions isolated environments, Fleets supplies hosted capacity, Spaces is an interactive desktop app for working alongside agents, and Cua Bench helps create and evaluate computer-use tasks. Choosing the right part starts with deciding whether you need to control an existing desktop, create a disposable computer, scale hosted workloads, collaborate at a desktop, or measure agent performance.
What is Cua?
Cua describes its SDK as a way to create and drive computers from code, with separate tools for desktop interaction, environment provisioning, hosted execution, collaboration and evaluation. The names refer to related components, not interchangeable products. Cua says its goal is an open-source, self-hostable layer for computer-use agents; that is the company’s positioning, not an independent assessment. Cua documentation · About Cua
As an Amazon Associate I earn from qualifying purchases.
Match the component to the job
| Component | Best fit |
|---|---|
| Cua Driver | Sending input to and inspecting windows in native desktop applications. |
| Sandbox | Creating isolated local computer environments, including Docker, QEMU and Apple Virtualization backends. |
| Fleets | Hosted, elastic capacity when a workload needs more concurrency than local machines can provide. |
| Spaces | Using an interactive desktop app to work alongside agents. |
| Cua Bench | Authoring computer-use tasks and running or scoring benchmarks. |
The repository also includes Lume, a local macOS virtual-machine tool, and CUA-S1 decision models. These are additional project components, not substitutes for the Driver, Sandbox, Fleets, Spaces or Bench. Cua repository
What does Cua Driver do?
Cua calls Driver an open-source computer-use driver for macOS, Windows and Linux. Through MCP or the CLI, an agent can target a chosen window to send input and inspect it. The product page says the driver can send input without moving the system cursor, but background input is not universal: some software requires the target window to be focused. Driver reports that constraint so a caller can retry or allow foreground input. Cua Driver
#1 Best Overall
- EMPOWER YOUR PASSIONS ELEVATE YOUR GAME – Whether you’re dominating the leaderboard, streaming your gameplay live, or tackling creative projects, the Lenovo Legion Tower 5i is an expandable powerhouse ready for anything.
- BEYOND FAST – The Intel Core Ultra 7 265F CPU is designed to give you the power boost you need to dominate the latest and most popular AAA games.
- GAME CHANGER – The NVIDIA GeForce RTX 5060 Ti GPU is beyond fast for gamers and creators. Experience lifelike virtual worlds, ultra-high FPS gaming, revolutionary new ways to create, and unprecedented workflow acceleration.
- BOLD DESIGN AND EFFORTLESS UPGRADE – The Legion Tower 5i’s transparent, tool-less side panel lets you easily upgrade and showcase your rig, while the customizable RGB lighting adds a personal touch to every session.
- FUTURE-PROOF YOUR PASSIONS – The Legion Tower 5i delivers stutter-free gameplay, fast loading times, and seamless multitasking. It’s equipped with 16GB and expandable to 128GB of 5600MHz DDR5 memory.
Three ways to run Driver
- MCP stdio server: connect an MCP client that can spawn a process.
- Long-running daemon: keep a driver process available for repeated interactions.
- One-shot CLI: issue an individual command from the command line.
Cua lists MCP clients such as Claude Code and Codex as examples. The Driver page states that Driver is MIT licensed; do not assume that this licensing statement covers every component in the broader repository, since optional perception extensions have separate licensing notices. Cua Driver · Cua repository
Documented platform labels
Cua’s Driver page lists macOS Sequoia and Tahoe, Windows 11 and Server 2025, and Linux X11 plus XWayland. These are the platform labels in Cua’s documentation, not a guarantee that every application, display server or configuration behaves identically. Check the live Driver page and Quickstart for current compatibility and installation details.
When to use a local macOS sandbox
The local macOS VM route is for people who need an isolated macOS environment on their own hardware. Cua’s guide specifies an Apple silicon Mac, macOS 14 or later, about 40 GB of free disk space and 8 GB of free memory. Those requirements apply to this local macOS path; they do not establish a hardware requirement for Driver, Linux or Windows work, or hosted Fleets. Create a local macOS sandbox with the Cua CLI
Rank #2
- EVOLUTION RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 128GB pool, which is perfect for running LLMs such as Deepseek 70B Q8, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 12% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
What the guide’s example involves
Cua’s tutorial demonstrates creating a Tahoe sandbox with its CLI. It reports a 23 GB initial image pull taking about eight minutes and subsequent clones ready in about 25 seconds. These are figures from Cua’s documented workflow, not independent benchmarks or guaranteed timings; actual results can vary with machine and network. The guide 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 all images have the same setup.
- Install the Cua CLI following the local macOS sandbox guide.
- Check for or install the Lume runtime as directed in the guide.
- Create a sandbox using the tutorial’s Tahoe example.
- Run a command in the sandbox, then capture a screenshot or open the desktop viewer.
- Remove the sandbox when it is no longer needed.
A Mac mini is one possible category of Apple silicon hardware, but Cua’s cited guide does not recommend a particular form factor or model. Buying a Mac is relevant only if you choose this local macOS VM path.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Local Sandbox or hosted Fleets?
Sandbox and Fleets address different deployment choices. Cua describes local Sandbox backends including Docker, QEMU and Apple Virtualization. Its hosted Fleets option is aimed at workloads needing more concurrency. Local execution gives you control over the machine and environment; hosted capacity avoids requiring all compute to come from your own local hardware. Consider application and OS needs, foreground behavior, integration method, expected concurrency and your cost model before choosing. Cua homepage · Cua documentation
Rank #3
- Built for Local AI Development: AMD Ryzen AI Halo is designed for local AI development and inference, featuring 128GB unified memory and support for up to 200B parameter models to build and run intensive AI workloads locally.
- 128GB Unified Memory: Features 128GB LPDDR5x unified memory at 8000 MT/s with 256 GB/s memory bandwidth, providing a shared memory pool across the CPU, GPU, and NPU to support larger AI models.
- AMD Ryzen AI Max+ 395 Processor: Features 16 cores, 32 threads, and Zen 5 architecture, paired with AMD Radeon 8060S integrated graphics featuring 40 RDNA 3.5 compute units and an AMD XDNA 2 NPU with up to 50 TOPS.
- Linux AI Developer Platform: Purpose-built for Linux-based AI development with full AMD ROCm software support and preloaded tools, models, and workflows optimized for local AI development.
- Compact, Connected Design: Includes a 2TB M.2 SSD, 10GbE LAN, Wi-Fi 7, Bluetooth 5.4, USB-C connectivity, and HDMI 2.1b.
Hosted Fleet rates
Cua’s homepage listed usage prices of $0.044625 per vCPU-hour and $0.0223125 per GiB-hour when accessed on October 7, 2026. These are Cua’s posted rates on that date, not a fixed total cost: usage and resource allocation determine the bill. Check the current Fleet pricing before estimating a deployment.
Where Spaces and Cua Bench fit
Spaces is Cua’s desktop app for working alongside agents, rather than the driver or the environment-provisioning layer. Cua Bench serves a different purpose: teams can author computer-use tasks and run or score benchmarks. If the question is whether an agent can complete a task reliably, benchmark authoring and evaluation matter; if the need is to operate an application, Driver or a provisioned environment is the relevant starting point. Cua documentation
Is Cua the right fit?
- Choose Driver when an agent must interact with a native desktop window, and verify the specific app’s focus and input behavior.
- Choose local Sandbox when you need an isolated environment on your own machine and meet the requirements for the selected backend.
- Consider Fleets when hosted execution and greater concurrency are more important than keeping compute local.
- Consider Spaces when people need an interactive desktop for working alongside an agent.
- Use Cua Bench when you need to build and evaluate computer-use tasks rather than simply run them.
Cua’s company page 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 background and ambition are claims published by Cua itself. About Cua
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.
