GuidesPrivacy & controlHistory, audio, and retention
Beginner4 min read

History, audio, and retention

InkSpoke keeps a local record of what you dictated — including a copy of the audio. Here is how to see it, prune it, and wipe it.

Every dictation is written to a local database so you can review, re-copy, and re-run past results. It never syncs and never leaves the machine — the History page carries a "Local only" badge for exactly this reason.

What is kept

  • The raw transcript, as the speech model heard it
  • The refined text, after the refinement model rewrote it
  • Timing and status metadata
  • A compressed copy of the audio — Opus in an Ogg container, 16 kHz mono
!
The audio copy is saved for every dictation and there is currently no switch to prevent it. If that is unacceptable for your threat model, set a short retention window and clear history regularly. (The Keep audio setting you may have seen belongs to the Meetings feature and does not govern dictation.)

Setting a retention window

SettingsGeneralHistory Retention

Retention is by age, not by item count. The options are: no limit, one day, one week, one month, three months, six months, and one year. The default is no limit — meaning history accumulates forever until you change it.

Pruning runs opportunistically after a new dictation is saved, not on a timer. If you set a short window and then stop dictating, old rows survive until the next dictation triggers a sweep. To delete immediately, use Clear history.

Deleting history

  • One entry — delete it from the History page. The matching audio file is deleted with it.
  • Everything — Settings → General → Clear history now. You will be asked to type a confirmation phrase, because this is not undoable.

There is no option to disable history recording altogether. The closest equivalent is a one-day retention window plus periodic clearing.

One exception worth knowing

History itself never uploads. But if you dictate using a cloud speech model, that recording is sent to the provider as part of transcription — that is the transcription request, not history sync. Long recordings that fail to upload are queued and retried, and the history row shows that state.

Next in this track

Keep going.