OpenHush

December 28, 2025 · View on GitHub

Local voice-to-text that just works—with the power to do more when you're ready.

Press a hotkey, speak, release — your words appear where your cursor is.

100% local. Your voice never leaves your device.

Privacy First

OpenHush processes everything on your machine:

No cloud — Transcription runs locally using Whisper AI
No account — No sign-up, no login, no tracking
No network — Works fully offline after initial model download
No data collection — We can't see your voice data because it never leaves your computer
Secure secrets — API keys stored in your system's keyring, never in plain text

Your voice is yours. Period.

Quick Start

# Install and run the wizard
openhush wizard

The wizard will:

Download the right model for your hardware
Configure your hotkey
Test your microphone
Start the daemon

That's it. Hold Right Ctrl and speak.

Or if you prefer manual setup:

openhush model download small   # Download a model
openhush start                  # Start the daemon

Capture Anything

OpenHush supports multiple audio sources:

Input	Description
Microphone	Any USB, built-in, or Bluetooth mic
System Audio	Transcribe meetings, calls, videos, podcasts playing on your machine
Multi-Channel	Select specific channels from pro audio interfaces
Wake Word	"Hey Computer" for hands-free activation—no hotkey needed

Mix and match. Capture your mic AND system audio together for meeting transcription with your own comments.

Output Anywhere

Route your transcriptions wherever you need them:

Destination	What It Does
Cursor	Paste directly into any application
Clipboard	Copy for manual paste
File	Save transcripts to disk (text, JSON, SRT)
Translation	Speak in any language, get English text
LLM Correction	Grammar, punctuation, filler word removal via Ollama or OpenAI
Meeting Summaries	AI-generated summaries of long recordings
Custom Hooks	Trigger any script—send to Notion, commit to git, post to Slack
REST API	Integrate with external tools, Stream Deck, Home Assistant

Output Pipeline Example

Voice → Noise Reduction → Whisper → LLM Cleanup → Save to File + Paste at Cursor

You control every step.

Features

Easy Mode (Works Out of the Box)

Push-to-talk — Hold hotkey, speak, release
Toggle mode — Press once to start, again to stop
Auto-paste — Text appears at your cursor automatically
System tray — Runs quietly in the background
Preferences GUI — Point-and-click settings

Power Mode (When You Need More)

Continuous dictation — Voice Activity Detection for natural pausing
Streaming transcription — See words appear as you speak
Noise reduction — RNNoise AI filters background noise in real-time
Filler word removal — Strips "um", "uh", "like" automatically
Custom vocabulary — Boost recognition for names, jargon, technical terms
Text snippets — Expand abbreviations (e.g., "sig" → your email signature)
Post-transcription hooks — Run shell commands after each transcription
REST API + Swagger UI — Full remote control and automation
D-Bus integration — Desktop integration on Linux

Quality Presets

Preset	Model	Best For
Instant	small	Quick notes, chat
Balanced	medium	Everyday dictation
Quality	large-v3	Important documents

Installation

Quick Install (Recommended)

# Arch Linux (AUR)
yay -S openhush

# Flatpak (coming soon)
flatpak install flathub org.openhush.OpenHush

# Homebrew (macOS)
brew install openhush

From Source

git clone https://github.com/claymore666/openhush.git
cd openhush
cargo build --release --features cuda  # or: hipblas, metal, vulkan
sudo cp target/release/openhush /usr/local/bin/

See Building from Source for GPU-specific instructions.

Usage

Everyday Commands

openhush start              # Start in background
openhush stop               # Stop the daemon
openhush status             # Check if running
openhush preferences        # Open settings GUI

Model Management

openhush model list                 # See available models
openhush model download medium      # Download a model
openhush model remove tiny          # Delete a model

File Transcription

openhush transcribe meeting.wav                    # Transcribe a file
openhush transcribe recording.mp3 --output json    # Output as JSON
openhush transcribe --summarize interview.wav      # Generate AI summary

Configuration

openhush config --show              # View current settings
openhush config --hotkey F12        # Change hotkey
openhush config --model large-v3    # Use most accurate model
openhush config --language de       # Set language (or "auto")

System Requirements

Component	Minimum	Recommended
GPU	None (CPU works)	NVIDIA, AMD, or Apple Silicon
RAM	4 GB	8 GB
Storage	500 MB	4 GB (for large models)

GPU Acceleration

GPU	Feature	Platform
NVIDIA	`cuda`	Linux, Windows
AMD	`hipblas`	Linux
Apple Silicon	`metal`	macOS
Any	`vulkan`	All platforms

Platform Support

Platform	Status	Notes
Linux (X11)	Full	Primary development platform
Linux (Wayland)	Full	KDE Plasma, GNOME, Sway
Linux (TTY)	Full	Terminal-only mode
macOS	Full	Intel & Apple Silicon
Windows	Full	Windows 10/11

Configuration

Settings live in ~/.config/openhush/config.toml:

[hotkey]
key = "ControlRight"
mode = "push_to_talk"       # or "toggle", "continuous"

[transcription]
preset = "balanced"         # instant, balanced, quality
language = "auto"           # or "en", "de", "es", etc.
translate = false           # true = always output English

[audio]
channels = "all"            # or [0, 1] for specific channels

[output]
clipboard = true
paste = true

[correction]
enabled = true              # LLM post-processing
remove_fillers = true       # Remove um, uh, like
ollama_url = "http://localhost:11434"
ollama_model = "llama3.2"

[wake_word]
enabled = false
phrase = "hey computer"

[appearance]
theme = "auto"              # auto, light, dark

Building from Source

git clone https://github.com/claymore666/openhush.git
cd openhush

# CPU only
cargo build --release

# With GPU acceleration
cargo build --release --features cuda      # NVIDIA
cargo build --release --features hipblas   # AMD
cargo build --release --features metal     # Apple Silicon
cargo build --release --features vulkan    # Cross-platform

Linux Dependencies

# Debian/Ubuntu
sudo apt install libasound2-dev libdbus-1-dev libpulse-dev pkg-config

# Fedora
sudo dnf install alsa-lib-devel dbus-devel pulseaudio-libs-devel

# Arch
sudo pacman -S alsa-lib dbus libpulse

Troubleshooting

Nothing happens when I press the hotkey

openhush status           # Is it running?
openhush start -f -v      # Run in foreground with verbose logs

Transcription is slow

Try a smaller model: openhush config --model small
Verify GPU is working: nvidia-smi or check logs

Where are the logs?

Main log: ~/.local/share/openhush/openhush.log
Crash reports: ~/.local/share/openhush/crash.log

Documentation

Visit the OpenHush Wiki for:

User Guide — Detailed usage
Architecture — System design
REST API — API reference
Hooks & Automation — Scripting guide

Contributing

Contributions welcome! See CONTRIBUTING.md for guidelines.

License

MIT License — see LICENSE for details.

OpenHush is not affiliated with OpenAI. Whisper is OpenAI's open-source speech recognition model.