Stillheard
privacy policy.

Effective date: 20 July 2026

Stillheard is a private, on-device app for keeping and hearing family voices. This policy is short because the honest answer to almost every privacy question about Stillheard is: the data never leaves your devices, and we never see it.

What we collect

Nothing. Stillheard has no accounts, no analytics, no advertising, no crash reporting service, and no server of ours. We do not collect, receive, store, sell, or share any personal information, and we could not: the app is designed so that your recordings, voices, and generated audio exist only on your own devices.

What stays on your devices

Voice recordings you import or record (including the spoken consent sentence), the voice profiles built from them, every generated audio keepsake, and your keepsake history. It also holds everything the later features make, on the same terms: transcripts of clips you choose to transcribe, the text of any book you import and the audiobooks made from it, the scripts and plays made from a story, and the small pronunciation list you teach the app. Alongside these live the provenance ledger and activity log, your device-approval list, and your settings. On the Mac these live in the app’s own folder (or a folder you choose); on iPhone, in the app’s private storage. Deleting the app, or the folder, deletes them. The family pairing secret is stored in the platform Keychain on each device and is not synced.

Network use, completely enumerated

Stillheard uses the network for two purposes, and neither one carries your content: downloading the machine-learning models it runs on, and, only if you turn it on, serving to your own devices.

The models are public downloads from Hugging Face’s servers, each a plain file transfer that sends none of your recordings, voices, or other content: the speech engine (about 1.4 GB) the first time the Mac app prepares to speak; a transcription model (about 1.6 GB) only if you choose to install it from the Models screen; and a small voice-isolation model that downloads the first time you use “Clean” on a recording while creating a voice. Every one of them runs entirely on your Mac once downloaded (transcription and voice isolation included), so your audio is never sent to any service.

Second, if (and only if) you turn on “Serve to family devices” in the Trust tab, the Mac app answers requests from devices on your own local network. Those requests must be signed with your family phrase and individually approved by you, per device, by name. Serving is off by default, the audio it produces is delivered over your local network only, and every request is written to the on-device activity log.

That is the whole of it: nothing is ever uploaded to us or to any third party, and no third-party AI service receives any of your data.

Microphone, camera, and local network permissions

The microphone is used for one thing: recording a consent sentence and fresh voice references when you create a voice, always started by you. The camera is used for one thing too: reading the pairing code shown on your Mac when you add a device. Nothing is photographed, saved, or sent; the code is read in memory and discarded, and you can pair without the camera by typing a code or the twelve words instead. The local-network permission lets the iPhone app discover and talk to your own Mac. Every permission does exactly what its system prompt says and nothing else.

Voice data is sensitive

A voice is biometric-adjacent data and Stillheard treats it that way: consent is required in the creation flow itself (a living person records their consent in their own voice; a memorial voice requires attested authority and a second family member), consent can be revoked in the app at any time and the voice stops being served immediately, every generated clip (a keepsake, a story, an audiobook chapter, a play) is marked as synthetic in a machine-readable way, and getting a single keepsake clip out of the app requires two people (a whole audiobook the family makes for itself saves directly, still marked and logged). These are product behaviors, not policy promises; the app enforces them.

The books that come with the app

Stillheard includes a small library of classic books to narrate. These are public-domain texts (their copyright has long expired), cleaned and packaged inside the app; using them involves no network and no third party. Any book you bring in yourself is read from the file you choose and stays on your device like everything else.

Children

Stillheard is a family app but is not directed at children. Creating a voice requires the consent flow described above; decisions about a child’s voice belong to their parent or guardian, and we recommend against creating voices of minors.

Changes and contact

If a future version ever changes any of the above (for example, if an optional sync feature is added), this policy will change first, visibly, with the app update notes. Questions: saras.app.support@gmail.com.