Privacy policy.
Your library is processed locally. Silo Systems collects none of it.
The short version
Silo Systems does not collect personal data through Silo Read.
There is no Silo account, analytics SDK, advertising identifier, telemetry, remote crash reporter, cloud AI endpoint, or document-processing service. Documents, imported audio, extracted text, reading and listening positions, collections, annotations, project edits, local answers, personal voices you import, and generated audio stay in the app sandbox or an export location you select.
Limited network activity is described below. In particular, optional Mac model downloads go directly to Hugging Face after separate consent, and that host receives ordinary web-request metadata such as your IP address. Silo Read never includes document content, prompts, annotations, or usage history in those model requests.
Local architecture
On macOS, the SwiftUI host communicates with its bundled processing runtime over an app-private Unix-domain socket inside the sandbox. It does not open a public HTTP server or accept connections from your network. On iPhone and iPad, the equivalent document, library, intelligence, and project APIs run in-process inside the app.
OCR, full-text indexing, speech, audio playback, audiobook assembly, transforms, retrieval, and citation validation run on the device. Apple Foundation Models are used only on supported devices and only through Apple's on-device framework. When a local provider is unavailable, Silo Read reports that state or returns an evidence-limited answer; it does not send the task to a cloud fallback.
Network activity
Silo Read can make these disclosed outbound requests:
- Optional Mac model downloads. After you review and approve the exact size, license, revision, storage requirement, and device fit, Silo Read can download Kokoro narration weights, the Gemma local text model, or the personal-voice model pair from Hugging Face. Downloads are atomic, resumable, cancellable, hash-verified, removable, and usable offline afterward. The reference audio of a voice you import is never uploaded and is never part of a model request.
- App Store purchase, update, and receipt traffic. Apple handles this traffic under your Apple Account. Silo Systems receives no payment-card information.
- Links you deliberately open. Privacy, support, license, model-card, and project links open in your browser or the appropriate system app, which then makes the request under its own settings.
Silo Read does not include iCloud library sync, proprietary cloud-drive account integrations, remote article extraction, or a production remote model-catalog refresh. Files from iCloud Drive, Dropbox, Google Drive, or OneDrive can be selected through Apple's Files interface when their provider is installed; that provider's own service terms apply before the selected file reaches Silo Read.
Optional model disclosures
- Kokoro-82M-bf16 — optional on Mac, bundled on iPhone and iPad; Apache-2.0; pinned Mac repository
mlx-community/Kokoro-82M-bf16at revisiona71e4d38b236d968966a2002c4c895dbd12b1c3c. - Gemma 4 E2B OptiQ 4-bit — optional on Mac; Gemma Terms of Use; pinned repository
mlx-community/gemma-4-e2b-it-qat-OptiQ-4bitat revisionb7a34a6d76bbebebe700ffe68b57c97a4e3c3622. - Chatterbox Multilingual: optional on Mac, for speaking in a personal voice imported from Silo Clone; MIT, an MLX conversion of Resemble AI's Chatterbox Multilingual; pinned repository
mlx-community/chatterbox-multilingual-v3at revision03565773edd72e949572557597af8063bb49a18a. - S3 speech tokenizer v2 — optional on Mac, the required companion encoder for personal voice; Apache-2.0 lineage from
FunAudioLLM/CosyVoice2-0.5B; pinned repositorymlx-community/S3TokenizerV2at revisione0c9886f0e1c35ae85b1f27277416fb19fc72bec.
The two personal-voice models install together as one consent-gated download and neither activates alone. Personal-voice synthesis runs entirely on your Mac.
Audio synthesized with a personal voice carries no watermark. The MLX runtime path does not embed Resemble's PerTh audio watermark, so nothing marks the output as synthetic. Silo Read documents that absence rather than claiming a protection it does not provide.
The iPhone and iPad app ships its supported Kokoro runtime assets inside the signed app. Apple system voices are enumerated from the operating system and require no Silo model download. Personal voice is a Mac-only capability.
Personal voices from Silo Clone
Silo Read does not record audio or create voices. On Mac, you can import a .silovoice file that Silo Clone exported. Silo Read reads only the file you choose in the open panel, refuses a file saved without Silo Clone's rights confirmation, converts its reference clip to a standard audio format, and stores that clip in the app container. Nothing about the voice is sent anywhere, and speaking in it happens on your Mac after the optional personal-voice models above are installed.
Files and permissions
- Files and folders you choose. Silo Read receives only files selected through Apple's document interfaces or folders explicitly granted on Mac. Security-scoped bookmarks preserve access where Apple permits it.
- Network client on Mac. Used for the optional model downloads and links described above. It is not used to transmit your library or generated content.
- Background audio on iPhone and iPad. Allows spoken text and imported audiobooks to continue while the screen is locked or another app is active.
Silo Read does not request Microphone, Camera, Photos, Contacts, Location, Health, advertising tracking, or unrestricted Full Disk Access. It never records audio.
Data stored on your device
Silo Read stores imported assets, canonical text, FTS indexes, collections, reading and listening positions, bookmarks, highlights, notes, project variants, render state, generated audio, reference clips of personal voices imported from Silo Clone, model files, and local logs in the platform app sandbox. Exports are written only to a destination you choose. Deleting an imported personal voice removes its reference audio from the container.
On iPhone or iPad, deleting the app rather than offloading it removes its sandbox data. On Mac, removing the application bundle can leave sandbox data behind; delete library items and optional models in the app first, then use macOS storage controls if you want the remaining container removed. Files exported elsewhere remain until you delete them.
Children's privacy
Silo Read is rated 4+. Silo Systems does not knowingly collect personal data from children or adults through the app; the same local-processing policy applies to every user.
Changes
If Silo Read materially changes its data practices, this page and the in-app disclosure will be updated before the new behavior is enabled. A feature that sends library content off-device would require a clear, separate opt-in; it will not be introduced through silent model or catalog updates.
Contact
Questions about this policy: admin@ohmslaw.net