Privacy policy.

Silo Fable collects no data. None.

No account, analytics, telemetry, advertising, tracking, crash-reporting SDK, Silo server, cloud transcript, or remote inference service is part of the app. Silo Systems cannot see your voice turns, character names, conversations, memory ledger, journal, imported voice, device identifiers, or usage.

Microphone audio stays in memory

Microphone permission is requested only when you set up voice input. While you hold the stage control, audio stays in memory for the speech engine you chose. Silo Fable never writes hold-to-talk audio to a file and does not use the microphone to record a voice. Release, cancel, interruption, failure, or backgrounding stops capture, and the audio is not kept. Earlier versions could also record a personal-voice reference clip of up to 15 seconds in Settings; that clip, and any voice data derived from it, stays in the app's private, backup-excluded storage until you delete it in Settings or delete the app, and is never uploaded.

Speech recognition stays on device

You choose a speech engine during setup and can change it in Settings. With Apple Speech, Silo Fable requires Apple's recognizer to report on-device support for the current language and marks every request as on-device only, refusing Apple's cloud fallback. With the optional downloadable speech model, transcription runs inside the app after you release the control, and Speech Recognition permission is not requested. An unavailable engine never falls back to the other engine or to a network recognizer.

Generated replies stay on device

On compatible iOS and iPadOS 26 devices, Apple's on-device model, built into the operating system, writes companion replies locally. Silo Fable uses no API key, backend, cloud model, or canned responder, and it never downloads a reply model of its own. Unsupported devices receive an explanation and a next step.

Voices from Silo Clone

Silo Fable does not record or create voices. You can import one .silovoice file exported by Silo Clone; a file saved without Silo Clone's rights confirmation is refused. Silo Fable reads only the file you pick and keeps a reference clip of up to 15 seconds from it, with its import record and derived data, in the app's private, backup-excluded storage until you replace or remove the voice in Settings. It is never uploaded, used for training, or shared with other apps. The file you exported from Silo Clone is not changed. A voice recorded by an earlier version of Silo Fable is no longer played, and Settings offers to remove it.

Optional model downloads

Silo Fable can make two downloads, each only after you review its disclosure and tap Download: a speech-recognition model of 1,017,761,038 bytes, if you choose the downloadable speech engine, and a voice model of 407,972,690 bytes, needed only to speak in a voice imported from Silo Clone. Both come from pinned revisions hosted on Hugging Face. Every file's size and checksum must match before the model is used, and staged files are excluded from backup. No audio, transcript, voice, name, or memory is sent with either request. Each model can be removed in Settings. If you never choose these options, neither download happens.

Your local archive

Your adult confirmation, character names, selected character, transcripts, replies, memory ledger, tide journal, and the number of free spoken replies used are stored as data inside the app's private Application Support container. The archive uses iOS Data Protection and is excluded from device backup. You can inspect, edit, pin, forget, or reset memory and remove journal entries in the app. The free-reply count is never copied to the Keychain or synced. Deleting the app removes its local container.

The Today's Tide widget reads one small shared file that holds only which character you selected. It never receives the names you chose, conversations, journal text, or memory.

Permissions

• Microphone: foreground hold-to-talk input. Earlier versions also used it to record a personal voice's short reference clip if you created one in Settings; voices now come from Silo Clone.
• Speech Recognition: only when you choose Apple Speech, for on-device transcription in the current language.
• Notifications: only if you turn on the daily reminder. The reminder has fixed wording scheduled on this device; no name, conversation, journal, or memory text appears in it.
• Files you choose: a .silovoice file you pick to import, and the location you pick for a journal export.

The app does not request contacts, photos, location, camera, tracking, Bluetooth discovery, or an account.

Outbound traffic

Silo Fable itself sends no personal data, and Silo Systems operates no server that receives anything from it. Network activity is limited to the optional model downloads described above, App Store purchase, entitlement, rating-prompt, and update traffic handled by Apple, availability of Apple's built-in model managed by the operating system, and external links you deliberately open. Hugging Face and its download hosts see each model download request itself, including your IP address, but no audio, transcript, voice, name, or memory. A browser request to a support, privacy, subscription-management, or crisis-resource page is handled outside Silo Fable under that site's policy.

Subscriptions

Apple processes purchases, renewals, cancellations, and verified entitlement status. Silo Fable receives signed StoreKit transaction state needed to unlock access; Silo Systems receives no payment-card information and maintains no customer account.

Adults and safety

Silo Fable is designed for adults and asks for an 18-or-older confirmation before onboarding. The characters are fictional entertainment software, not therapists or crisis services. Settings links to external support resources; opening one leaves the app.

Changes

If a future version materially changes this policy, this page and the in-app disclosure will be updated before that behavior ships. The policy remains: Silo Systems collects no data.

Contact

Privacy questions: admin@ohmslaw.net.