trycua/cua at 17K Stars: Open-Source Infrastructure for Computer-Use Agents Ships Sandboxes, SDKs, and Benchmarks for macOS, Linux, and Windows

GitHub·medium signal

CUA provides the infrastructure layer for training and evaluating AI agents that can control full desktop environments across macOS, Linux, and Windows. The project ships sandboxes for safe agent execution, SDKs for building computer-use capabilities, and benchmarks for comparing agent performance on real desktop tasks. At 17K stars, it fills a critical gap between browser-only automation tools and full OS-level agent control, positioning itself as the testing ground for the next generation of autonomous desktop agents.