Privacy

On-device means on-device.

Cloud dictation is a product decision dressed up as a feature. VV’s default path is a local model on your own Mac or Windows PC. The audio you just spoke is not a training sample for someone else.

The default English model is NVIDIA Parakeet, running on your computer via ONNX. Voice activity detection uses Silero. Optional Whisper-family models are Apache-2.0. None of that requires a VV login.

Transforms are separate. They can stay on your computer: Apple Intelligence on a Mac, or a local Ollama model on a Mac or PC. If you point one at OpenAI, Anthropic or Groq, that Transform says so on its face. Standard dictation does not follow it out the door. Pro’s Cloud Polish and meeting summaries are the same kind of choice: they send text to Vaitly’s server, which uses OpenAI, never the audio.