trycua/cua: YC-Backed 'Docker for Computer-Use Agents' Hits 13K Stars — macOS/Linux/Windows Desktop Sandboxes at 97% Native CPU Speed

GitHub·high signal

CUA provides open-source sandboxes, SDKs, and benchmarks for training and evaluating AI agents that control full desktops, running at 97% native CPU speed on Apple Silicon via lightweight virtual containers with H.265 video, shared clipboard, and audio passthrough. The project has matured from infrastructure library into a managed cloud platform with credit-based billing, positioning it as the Anthropic-agnostic substrate for any computer-use agent regardless of model. YC-backed (Summer 2025 batch) and actively developed through March 2026, it directly competes with Anthropic's own computer-use primitives.