According to 1M AI News monitoring, open-source developer Tommaso De Rossi released a desktop automation command-line tool called usecomputer, written in Zig as a native binary that does not depend on the Node.js runtime. It allows AI programming assistants (such as Claude Code, Codex, OpenCode, etc.) to directly control the desktop’s mouse, keyboard, and screenshots. It supports macOS and Linux (X11, Wayland via XWayland).
usecomputer provides CLI commands for screenshots, mouse movement/click/drag/scroll, keyboard input, and hotkey composition, along with a coordinate mapping system (coord-map) that automatically converts pixel coordinates in screenshots to actual screen coordinates. Screenshot output defaults to scaling the longest edge to 1568 pixels to fit the model context window. The tool also supports Kitty Graphics Protocol; after setting environment variables, screenshots can be directly embedded into the model context without additional file reading.
De Rossi previously developed the browser automation tool Playwriter (over 3,200 GitHub stars). usecomputer extends browser automation to desktop automation. The project is extracted from his single repository kimaki and offers both CLI and Node.js library usage options. The README includes complete integration examples for OpenAI’s computer tool and Anthropic’s computer use.