LumaSubs generates live, translated captions for whatever's playing on your Mac. This page explains, in full, what that requires and what it doesn't — there's no shorter version hiding behind it.
LumaSubs has no server, no user account, and no analytics. Every model, every transcription, and every translation runs on your Mac. The only network traffic LumaSubs ever initiates is a one-time download of a speech-recognition model, and that model is fetched from Hugging Face's public model repository — never through a server we operate. We could not see what you've captioned even if we wanted to.
The rest of this document is the detail behind that sentence.
LumaSubs reads audio from a source you pick, then transcribes it, translates it into your chosen subtitle language, and displays both as a floating caption overlay. Two kinds of source are available:
LumaSubs cannot capture from either source without the corresponding explicit, revocable macOS permission, and it does not access your camera under any circumstance.
If you use a call-audio source (FaceTime, Zoom, Teams, or a phone call via iPhone Mirroring), you are transcribing and, if History is on, storing the other participant's speech locally on your Mac. That's between you and the people you talk to — consider letting them know, and be aware that some places legally require the other party's consent to record or transcribe a call.
Captured audio — from either source — is held in memory just long enough to transcribe it, then discarded. LumaSubs never writes raw audio to disk and never transmits it anywhere. Transcription runs on Whisper: an open-source Core ML speech model that runs entirely on your Mac via Apple Silicon's Neural Engine. The model file itself (not your audio) is downloaded once from Hugging Face's public repository and then reused from disk on every later launch — no further network access.
Subtitle translation uses Apple's built-in Translation framework. Per Apple's own documentation, translation runs on-device once the relevant language pair is downloaded; Apple may log operational metadata such as which language pair was requested, but LumaSubs never sends the caption text itself anywhere outside Apple's on-device translation pipeline. See Apple's Privacy Policy for how Apple's own frameworks handle this.
Everything LumaSubs keeps stays in your Mac's local, sandboxed app storage — never in iCloud, never synced, never uploaded:
Uninstalling LumaSubs removes its app container, including all saved history and preferences.
LumaSubs's only outbound network use is downloading the Whisper model file from Hugging Face on first use of that model. There is no LumaSubs server. No usage statistics, crash reports, or diagnostics are ever transmitted by LumaSubs itself.
To be explicit, LumaSubs has no mechanism to collect any of the following, because it has nowhere to send them even if it tried:
You can turn history off entirely with the Save History toggle before starting captions. Existing sessions can be renamed, deleted individually, or cleared all at once from the History tab. Screen & System Audio Recording, Microphone, and any other system permission can be revoked at any time in System Settings → Privacy & Security.
LumaSubs is not directed at children and does not knowingly collect information from anyone, of any age — there is simply no collection mechanism to distinguish between users in the first place.
If LumaSubs's data handling ever changes, this page will be updated and the effective date above will change accordingly. Material changes will also be noted in the app's release notes.
Questions about this policy or how LumaSubs handles data can go to contact@pinenode.com.