●●●
[github] [releases]

$ cat README.md

keyless █

Privacy-first, real-time speech-to-text dictation

Local-only voice transcription built in Rust. Hold a hotkey, speak, release - your words appear as text.

[rust][MIT license][100% local][99 languages]

$ ./demos

▸ Desktop App (Beta)

Desktop app demo

▸ TUI (Terminal) Version

TUI demo

$ ./quick-start

▸ Desktop App (Beta)

Download desktop app · Releases

Look for keyless-desktop-* assets

# Download from releases page
# Install .dmg (macOS), .msi (Windows), or .deb/.rpm (Linux)

# Build from source

Prerequisites: Node.js (LTS), pnpm (v10+), Rust, Tauri CLI

git clone https://github.com/hate/keyless.git
cd keyless/keyless-desktop
pnpm install
pnpm tauri dev    # Development mode
pnpm tauri build  # Production build

First run: Grant microphone and accessibility permissions when prompted. Download a Whisper model from the Models screen. Configure your hotkey and output mode in Settings.

▸ TUI (Terminal) Version

Download binary · Releases

Look for keyless-* assets (not desktop)

# Download from releases page
# Extract and run
./keyless  # macOS/Linux
keyless.exe # Windows

# Install via Cargo (from git)

# Latest main
cargo install --git https://github.com/hate/keyless --package keyless --locked

# Or pin a release tag
cargo install --git https://github.com/hate/keyless --tag tui-v0.3.0 --package keyless --locked

# Update existing install
cargo install --git https://github.com/hate/keyless --package keyless --locked --force

# Enable CUDA (Linux/Windows with NVIDIA)
cargo install --git https://github.com/hate/keyless --package keyless --locked --features cuda

Note: CUDA is optional; add --features cuda on NVIDIA systems with the CUDA toolkit. macOS Metal is automatic.

# Build from source

git clone https://github.com/hate/keyless.git
cd keyless
cargo build --release
./target/release/keyless

First run: Downloads Whisper model. Usage: Press Control+Option (configurable) to start dictating.

$ ls features/

▸ Privacy-First Architecture

  • • 100% local processing
  • • No cloud, no API keys
  • • All data stays on device
  • • Open-source & auditable

▸ Real-Time Transcription

  • • Live preview while speaking
  • • Push-to-talk control
  • • Computed final transcription
  • • Multilingual (99 languages)

▸ Smart Quality

  • • No-speech detection
  • • Temperature fallback decoding
  • • Auto language detection
  • • Quality metrics tracking

▸ Performance Optimized

  • • GPU acceleration (Metal/CUDA)
  • • Quantized model support
  • • Auto mic rate (device default; prefers 48 kHz; caps high rates)
  • • High-quality resampling (rubato, FFT‑based)

$ cat motivation.txt

Built to create a fast, private dictation workflow using the fewest dependencies possible, entirely in Rust. Audio capture, ML inference, concurrency orchestration, and both a responsive terminal UI and modern desktop app—all local, no accounts required.