Cua

Agent-enablingFreemium

Background desktop control and cross-OS sandboxes for computer-use agents

Overview

About Cua

Cua combines a background driver for native applications with isolated desktops for computer-use agents. The open-source driver connects over MCP or CLI on macOS, Windows, and Linux, while Cua Sandbox and hosted Fleet cover environments across Linux, Windows, macOS, and Android.

So agents can…

  • Send clicks, typing, and scrolling to a chosen window without taking over the system cursor
  • Run shell or Python code and GUI actions inside the same isolated computer
  • Execute training, evaluation, and data-generation workloads in parallel across local sandboxes or hosted fleets

Listing decision

Why Cua is listed

Cua supplies computer-use drivers, sandboxes, fleets, and evaluation infrastructure that agents consume; it enables agent execution rather than acting as the autonomous agent.

Before you choose

Fit and limitations

Best for

Teams building or evaluating computer-use agents that need background native-app control, isolated desktops, or parallel runs across multiple operating systems.

Known limitations

Background input is not universal: some applications require foreground focus. Local sandboxes consume operator CPU, memory, disk, and compatible virtualization; hosted Fleet capacity is separately billed. Cua’s current sandbox documentation also says snapshot and fork operations are not implemented for local sandboxes or Fleet creation paths in cua-sandbox 0.7.0.

Sources

Sources and verification

First-party unless marked otherwise. Each link supports only its adjacent claim.

  1. Cua Driver

    Cua documents background window-targeted input on macOS, Windows, and Linux through MCP, a daemon, or one-shot CLI commands, with foreground fallback for software that requires focus.

    official product page · accessed Sep 26, 2026
  2. How Cua sandboxes work

    The documentation describes isolated computers that combine code and GUI control, local and Fleet execution, supported runtimes, lifecycle trade-offs, and current snapshot limitations.

    official documentation · accessed Sep 26, 2026
  3. Cua product and pricing overview

    Cua documents computer-use workloads across Linux, Windows, macOS, and Android, plus separately priced hosted Fleet capacity for parallel training, evaluation, and data generation.

    official product page · accessed Sep 26, 2026
  4. Cua repository license

    The linked Cua repository is licensed under the MIT License.

    official license · accessed Sep 26, 2026

Related tools

Comments