AI Briefingcua: Open-source stack providing real desktop environments for AI agents
trycua/cua
About the project
Provides the infrastructure needed when AI agents manipulate actual GUIs beyond code or APIs. Cua unifies isolated cloud desktops, local macOS VMs, and benchmark tools for evaluating agents' computer-use capabilities. Developers can connect their models and agents to perform complex desktop tasks.
Cua Driver provides tools to control native apps and browsers on macOS, Windows, and Linux. It features the ability for agents to perform tasks in the background without moving the mouse pointer or stealing screen focus. It integrates with various agents via CLI, MCP, and typed SDKs.
On Apple Silicon-based Macs, Lume leverages Virtualization.Framework to create and manage local VMs. Cua Fleets provisions isolated Linux desktops in the cloud, supporting command execution and screenshot saving. These environments are controlled through a consistent interface via the Sandbox SDK.
Cua Bench is used to build computer-use tasks, evaluate agents, and export training trajectory data. It allows for fast execution and verification of simulated tasks without VMs or Docker. Released under the MIT license, it provides an essential toolkit for computer-use agent development and research.
trycua/cua
Scale computer-use 2.0 with open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation.
HTML
This introduction was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report errors, attribution issues, or removal requests via Contact.