Voice Input (100% Local)¶
Hostess lets you dictate commands, commit messages, or prompts straight into the terminal. Zero audio ever leaves your PC — speech recognition runs locally via whisper.cpp with automatic Vulkan GPU acceleration.
Setup in 30 seconds¶
- Open Settings → Voice input.
- Optionally pick a fixed Language (such as English or Deutsch instead of Detect automatically — this skips language detection and roughly doubles recognition speed).
- Download the recommended model (downloads support pause/resume and are SHA-256 verified):
| Tier | Size | Recommendation |
|---|---|---|
Small (base) |
~148 MB | Older CPUs without a dedicated GPU |
Medium (small) |
~488 MB | Balanced choice for modern multi-core CPUs |
Large (turbo-q5) |
~574 MB | High accuracy, under a second with a discrete GPU |
Maximum (turbo) |
~1.6 GB | Best accuracy — recommended with a dedicated GPU |
- Use Test microphone at the bottom to verify input levels before dictating.
Dictating into the terminal¶
- Click the microphone icon in the header or press Ctrl+Shift+M.
- As you speak, the button ring pulses with your voice level and recognized words stream live to your cursor.
- Never presses Enter: Hostess never submits a command automatically, so you can always review the text first.
- For safety, recording stops automatically after 10 minutes.