S U P E R V O I C E M O D E
Local voice dictation for macOS.
Nothing leaves your machine.
[ demo video — TBC ]
What it does
- Hold a key, speak, release. Text appears at your cursor.
- Runs Parakeet or Whisper STT entirely on-device.
- Local LLM fixes grammar, punctuation, and technical terms — no API keys needed.
- Code-aware: detects your editor, language, and file. Applies correct naming conventions.
- 80+ built-in developer term corrections. Extensible with your own rules.
- Voice assistant with Claude, Codex, Ollama, or a local LLM. Speaks answers back.
What it doesn't do
- Send your audio or text anywhere.
- Require an account, login, or API key.
- Need an internet connection.
Before / After
Raw STT output vs. what Super Voice Mode produces.
Raw STT
so um import the num pie library and create a data frame
Output
import the numpy library and create a DataFrame
Raw STT
add a function called validate token
Output
add a function called validate_token
Raw STT
so um we need to do three things first update the database second run migrations and third restart the server
Output
We need to do three things:
1. Update the database
2. Run migrations
3. Restart the server
1. Update the database
2. Run migrations
3. Restart the server
Raw STT
the function returns true if x is greater than zero
Output
the function returns true if x is greater than zero
How it works
Audio Capture
↓
Voice Activity Detection (Silero VAD)
↓
Speech-to-Text (Parakeet / Whisper — on-device)
↓
Deterministic Replacements (fillers, tech terms, numbers)
↓
LLM Correction (Qwen 1.7B — on-device, optional)
↓
Clipboard → Paste
Every step runs locally. The entire pipeline completes in under 2 seconds.
Pricing
Free during beta.
Pro pricing TBC.
Download
macOS 14+ · Apple Silicon · v0.4.6