Privacy Policy
The short version
We do not collect anything. Not your prompts, not your chats, not your recordings, not your model usage, not crash reports, not analytics events, not your IP address, not an advertising identifier. There is no account to create, no email to give us, no telemetry pipeline.
The app runs entirely on your device, inside Apple's standard per-app sandbox. The only outbound network calls are HTTPS downloads of public open-weight model files from the registry shown to you before each download.
No third-party AI providers. All language-model
inference happens locally on your device via the open-source
llama.cpp library. We do not send your prompts
or completions to OpenAI, Anthropic, Google, Microsoft, Amazon, or
any other AI service. We do not use any third-party AI API.
What data the app touches
| Data | Where it lives | Who can see it |
|---|---|---|
| Your prompts & chat history | Local app sandbox | Only you, on this device |
| Tape recordings | Local app sandbox (Documents/tapes/) | Only you, on this device |
| Downloaded model files (GGUF) | Local app sandbox | Only you, on this device |
| API tokens for the local server | Local app sandbox | Only you, on this device |
| App preferences & eval start date | Local UserDefaults | Only you, on this device |
| StoreKit purchase receipts | Local; verified on-device | You and Apple (App Store) |
Uninstalling the app deletes every byte. There is no copy anywhere else because there is no anywhere else.
Data we collect
None. We do not operate a server. There is nothing for us to collect.
Data we share with third parties
None. Specifically, we do not share data with:
- Analytics providers (no Google Analytics, no Firebase, no Mixpanel, no PostHog, no anything)
- Advertising networks (no AdMob, no AppLovin, no Unity Ads, no anything)
- Crash reporters (no Crashlytics, no Sentry, no Bugsnag)
- Third-party AI providers (no OpenAI, no Anthropic, no Google AI, no Microsoft AI — all inference is local)
- Attribution / install tracking SDKs
- Any parent, subsidiary, or related entity (we do not have any)
Network calls
- Model downloads. Standard HTTPS from the public
registries shown in the in-app catalog (e.g.
huggingface.co). These requests are sent by Apple's URLSession; the registries see a normal HTTPS request from your IP address, the same as any web download. We do not see these requests. - Local server (optional, off by default). An HTTP server you can enable in Settings. It binds to your device's interfaces (loopback only by default after first toggle, or your local Wi-Fi if you opt in). Every request requires a bearer token that only you generate. Traffic never leaves your local network.
The app does not phone home. There is no analytics endpoint, no remote config, no opt-in/opt-out telemetry switch — because there is nothing to opt out of.
Permissions
The app declares no special permissions: no microphone, no camera, no location, no contacts, no photos, no HealthKit, no Bluetooth, no push notifications, no background modes. Its entitlements file is empty.
Data retention & deletion
Because we do not collect any data, there is nothing for us to retain. To delete everything Vinyl LLM Studio knows about you:
- Long-press the Vinyl LLM Studio icon on your iOS home screen
- Tap Remove App → Delete App
That removes every byte: chats, recordings, downloaded models, preferences, purchase receipts. There is no copy elsewhere because we do not operate a server.
Children's privacy
The app is not directed at children under 13. Content guardrails block explicit, illegal, harmful, and copyrighted output by default. The model catalog is curated; users cannot side-load arbitrary weights through the UI.
Your rights (GDPR / CCPA / and analogues)
Because we do not collect or process your personal data:
- Right to access: already satisfied — there is nothing we hold.
- Right to delete / be forgotten: uninstall the app.
- Right to data portability: already satisfied — we hold nothing.
- Right to opt out of sale: already satisfied — we sell nothing.
- Right to withdraw consent: already satisfied — we never asked for any.
To file a formal request anyway, email phantotype@gmail.com with the subject line "Privacy request". We will respond within 30 days.
Third-party software components (full list)
The app links the following open-source libraries. These run on your device only; none of them phone home.
- llama.cpp (MIT) — on-device LLM inference
- FlyingFox (MIT) — embedded HTTP server (optional, off by default)
Open-weight model files (Llama, Mistral, Phi, Qwen, Granite) are downloaded at your request from public model registries; their use is governed by each model's published license.
Changes to this policy
We will update the "Last updated" date at the top of this document and publish the change in the app's release notes.