Cua

Cua provides infrastructure for running computer-use agents on Linux, Windows, macOS, and Android machines. It addresses the need to execute agent training, evaluation, and data-generation workloads across desktop and mobile environments, supporting tasks such as clicking, typing, scrolling, and inspecting accessibility trees without moving the system cursor or stealing focus.

The platform includes Cua Driver, an open-source MIT-licensed background driver that runs as an MCP stdio server, long-running daemon, or one-shot shell command on macOS, Windows, and Linux. Cua Sandbox provisions disposable GUI environments locally through Docker, QEMU, and Apple VZ, or through Cua Cloud. Cua Fleets manages warm pools of pre-booted machines for parallel rollouts, while Cua Bench supports eval and gym authoring. Lume manages macOS VMs on Apple Silicon. Integrations include OpenAI, Qwen Code, and Factory Droid. The platform also offers verified trajectory datasets with evaluators and human-reviewed annotations.

Cua is aimed at teams running computer-use agent training, evaluation, and data-generation workloads. Cua Sandbox is open-source and self-hostable; Cua Fleet uses usage-based pricing for CPU and memory. SOC 2 Type I has been completed, with Type II in progress, and the SOC 2 Type I report is available to customers and qualified prospects on request.