goodroot/hyprwhspr v1.22.0 on GitHub

Features

Model unload / reload

Free GPU VRAM without stopping the service — useful before running a local LLM or other GPU-intensive application.

hyprwhspr model unload   # Unload model from GPU memory (service stays alive)
hyprwhspr model reload   # Reload model back into memory

Recording is blocked while the model is unloaded, with a desktop notification on attempt
Tray shows 󰒲 icon in unloaded state when the signal file is present
Only applies to local backends (pywhispercpp, faster-whisper, onnx-asr); no-op for rest-api / realtime-ws
Tip: bind to keys in ~/.config/hypr/hyprland.conf:

  bindd = SUPER ALT, U, Unload Whisper model, exec, hyprwhspr model unload
  bindd = SUPER ALT, L, Reload Whisper model, exec, hyprwhspr model reload

Direct-type injection (inject_mode)

Adds a new inject_mode config option for terminals that use the Kitty keyboard protocol (Ghostty, Kitty, WezTerm), where synthetic Ctrl+V keystrokes from ydotool are misinterpreted.

{ "inject_mode": "wtype" }

Value	Behavior
`null` (default)	Clipboard + paste keystroke — existing behavior
`"wtype"`	Direct typing via `wtype` — works in all Wayland windows including Kitty-protocol terminals
`"ydotool_type"`	Direct typing via `ydotool type` — also works in Kitty-protocol terminals

Both modes bypass the clipboard entirely and respect auto_submit.

Gracefully falls back to clipboard+paste if the configured tool is not installed.

Full Changelog: v1.21.0...v1.22.0