The short version
Silo Transcribe collects no data. None.
No analytics. No telemetry. No account. No personal data. No usage metrics. No error beacons. No advertising IDs. No fingerprinting. No crash reports. No data of any kind is collected, stored, transmitted, or shared by Silo Transcribe about you, your device, or your conversations.
Everything Silo Transcribe does — recording, transcription, speaker diarization, summarization, and search — runs entirely on your Mac, on-device, using a bundled local runtime and locally-stored open-weights models (a Whisper speech-to-text model and a Gemma language model). There is no cloud component. There is no server. There is no Silo Transcribe account.
What we don't have, and therefore never see
Because Silo Transcribe has no server-side component, the following do not exist anywhere in our infrastructure (and our infrastructure consists of: nothing):
- Your name, email, phone, or address.
- Your IP address, device identifier, MAC address, or any other hardware fingerprint.
- The audio you record, the transcripts Silo Transcribe produces, the summaries it generates, or any file you export.
- The contents of your Documents, Downloads, Desktop, or any other folder.
- Session timing, feature usage, button clicks, or any other behavioural data.
- Crash reports. Silo Transcribe does not implement a crash-reporting SDK.
We never see any of this because none of it is ever sent. The app does not phone home, ever.
The local backend is not a network service
Silo Transcribe runs a small local Python process to perform AI inference. It binds strictly to the loopback interface (127.0.0.1), never to a public or network-facing address. This is on-device inter-process communication (IPC) only — a private channel between the macOS capture layer and the on-device transcription and summarization engines. It accepts no external connections and is not reachable from your network or the internet.
Outbound traffic you initiate
Three classes of outbound network traffic can originate from your Mac while Silo Transcribe is running. None of them go to Silo Transcribe. All of them are user-initiated:
- Model weight downloads. The first time you transcribe or summarize, Silo Transcribe downloads the relevant Whisper and Gemma model weights from public open-weights repositories — primarily Hugging Face. The download is anonymous by default; no token, account, or registration is required. Once the weights are on disk, every transcription and summary after that is fully offline.
- Mac App Store updates.When Apple ships a Silo Transcribe update, the Mac App Store delivers it. This traffic is between your Mac and Apple's servers, governed by Apple's privacy policy. Silo Transcribe is not involved.
- Links you click. If you click an external link inside the app (for example, opening this privacy policy), your default browser handles the request. Silo Transcribe is not involved.
Permissions we request, and why
Silo Transcribe asks for two macOS permissions. Both are used only to capture audio on your device for local transcription, and neither sends anything off your Mac:
- Microphone. To record your voice and others in the room through your input device.
- Screen Recording. macOS routes system audio capture through
ScreenCaptureKit, which is gated behind the Screen Recording permission. Silo Transcribe uses it to capture the system audio of a call — it does not record, store, or transmit your screen.
On-device data you create
The audio you record, the transcripts and summaries Silo Transcribe produces, and your local preferences live on your disk under ~/Library/Containers/com.silosystems.silotranscribe/Data/Library/Application Support/SiloTranscribe/ (plus any folder you explicitly choose to export to). Transcripts are indexed in a local SQLite database on the same disk so you can search them. They are yours. We do not have access to them. Deleting Silo Transcribe removes the app; your work remains under its container until you delete it manually.
Children's privacy
Silo Transcribe is rated 4+ on the Mac App Store. Because no personal data is collected from any user, no special treatment is required for users under 13. Same policy applies regardless of age: nothing collected.
Third-party services
Silo Transcribe uses these open-source components and public services. None of them receive any data about you from Silo Transcribe:
- Whisper (open-weights speech-to-text model, MLX port) — runs locally.
- Gemma (open-weights instruction-tuned language model) — runs locally via mlx-lm.
- MLX(Apple's on-device array/ML framework) — runs locally.
- Hugging Face Hub — contacted only when you transcribe or summarize for the first time and the model weights are downloaded.
Changes
If we ever materially change how Silo Transcribe handles data — for example, if a future version of the app ever begins to collect anything at all — we will publish an updated version of this page, announce the change prominently in the release notes, and require explicit consent in-app before any new collection begins. The effective date at the top of this document reflects the current policy.
The current policy is: nothing is collected.
Contact
Questions about this policy: admin@ohmslaw.net
For non-private inquiries you can also open an issue at github.com/flowmar47/silo-transcribe.