Privacy in KeyScribe.

Speech recognition is always local. Network use is limited to model downloads and optional BYOK rewrite calls for modes you enable.

Audio stays local

There is no cloud speech-to-text mode. Your microphone audio is transcribed on your Mac.

No telemetry

KeyScribe does not collect speech, transcripts, usage data, diagnostics, or crash reports.

Optional rewrite

Text can leave your Mac only when a mode uses a provider or endpoint you configured.

The boundary is explicit.

Privacy mode disables context sharing. Best-effort redaction can hide recognizable sensitive patterns before optional rewrite, but it is not a guarantee.

Saved API keys live in macOS Keychain.
History is local JSONL under Application Support.
Redaction maps live in memory only when a recognizable span is found.
Redaction depends on what speech recognition produced, so it can miss sensitive content.
For content that must never reach a provider, use Plain Dictation or another no-rewrite mode.

Read the full privacy document