Privacy on Device

What is local#

• The recording captured while you hold the key.
• The transcription itself, run by whisper.cpp on your hardware.
• The optional cleanup pass, which uses an additional local model and never sends text to a hosted AI service.
• Your personal dictionary of technical terms.

What uses the network#

• Downloading the speech model, once.
• Purchase and license activation.
• Update checks.

None of those carry your dictation. Once the model is installed, dictation itself works offline.

Why this is the deciding factor for some people#

Cloud dictation is a fine trade for a shopping list. It is a different question when the sentence you are about to speak contains a customer's name, an unreleased product detail, a private repository's architecture, or a command you are about to run against production.

With a hosted transcriber, the answer to "where did that audio go" is a policy document. Here it is a property of how the app is built: the audio is transcribed by a process on your machine and there is no upload step for it to travel through.

What this does not claim#

It does not claim the app never uses the network at all, because it does, for the three things listed above. The claim is narrower and checkable: your voice and the text it becomes are not part of that traffic.

Next#

• Local Transcription — the mechanism behind the claim
• FAQ — the short answer version
• Back to the documentation home