Look, then speak
Two stable cameras build a stereo gaze estimate. Say “click send” and the click lands where you were looking.
Gaze + voice for the Linux desktop
Look at a target, say what you mean, and Keyboard-Killer does the rest. Stereo cameras and local speech recognition stay on your machine.
curl -fsSL https://keyboard-killer.ohrats.party/install.sh | sh
curl -fLo /tmp/keyboard-killer.deb https://github.com/OhRats-Technologies/keyboard-killer/releases/latest/download/keyboard-killer_amd64.deb && sudo apt install /tmp/keyboard-killer.deb
The .deb includes the /dev/uinput udev rule, a desktop entry, and an optional systemd user unit.
Two cameras are required for stereo gaze; the app reports a single camera as degraded.
macOS is not available yet.
The macOS host is a capability stub in this release: CGEvent, AVFoundation, ScreenCaptureKit, TCC, and notarization are not implemented. No installer is published.
Read the platform matrixWindows is not available yet.
The Windows host is a capability stub in this release: SendInput, Media Foundation, Windows.Graphics.Capture, UI Automation, and signing are not implemented. No installer is published.
Read the platform matrixTwo stable cameras build a stereo gaze estimate. Say “click send” and the click lands where you were looking.
Parakeet speech recognition runs on your CPU. Camera, speech, and history stay on the machine.
Optional OpenAI-compatible, Ollama, or ACP agent providers. The deterministic pointer and voice paths never depend on one.
doctor and the setup window report cameras, audio, permissions, models, and providers with one next action.
Privacy
Screenshots only leave the device if you explicitly enable a remote gaze VLM. API keys live in the OS keychain.