Hush

Local LLM-powered speech-to-text. Press a hotkey, speak, and text lands wherever your cursor is. Fully offline — nothing leaves your machine.

How it works

1. Press Cmd+Shift+H (macOS) or Ctrl+Shift+H (Windows/Linux).
2. An overlay appears — speak.
3. Press the hotkey again (or pause for 1.5s) to stop.
4. Text is pasted directly at your cursor.
5. Every transcription is saved to ~/hush/history.txt.

Setup

# 1. Install dependencies
pip install -r requirements.txt

# 2. Run
python3 hush.py

The first run downloads the tiny Whisper model (~150 MB) from HuggingFace — one time only, no network needed after that.

Configure

Edit config.json:

Key	Default	Description
hotkey	<cmd>+<shift>+h	Toggle hotkey (pynput format)
model_size	tiny	Whisper model: tiny, base, small, medium
save_dir	~/hush	Where transcriptions are saved
max_history	500	Max lines in history file
stop_after_silence_sec	1.5	Auto-stop after silence
overlay_duration_ms	2000	How long the overlay stays after done

Requirements

• Python 3.9+
• A microphone
• macOS, Windows, or Linux