Privacy Policy

Hush collects nothing. This page exists to say that in detail.

Effective date

2026-05-18

App

Hush for visionOS

Developer

Juergen Klaassen / Klaassen Consulting

What Hush collects

Nothing.

Hush does not have a server. There is no account to create, no login, and no telemetry. Every conversation you have with the model happens on your Apple Vision Pro and never leaves your device.

No third-party AI services

Hush does not use any third-party AI service. All AI processing — reading your prompt, generating the response, every token in between — runs entirely on your Apple Vision Pro using Apple’s MLX framework. There is no remote inference, no cloud API call, and no external server involved in producing responses.

The AI model itself is software that runs locally on your device, the same way a word processor or a calculator runs locally. Your conversations are processed by code executing on your hardware, not transmitted to anyone.

The only data Hush stores locally on your device is:

• Your conversation history (turns of user input + model responses), persisted via Apple’s SwiftData framework. Stored exclusively in the app’s container; never transmitted anywhere.
• Your settings (chosen theme, voice output toggle, thinking-mode toggle), persisted via Apple’s UserDefaults. Never transmitted anywhere.

Tapping “Reset conversation” in the app’s Settings sheet deletes all stored conversation history immediately and permanently.

What Hush transmits

Exactly one network request, and only on first launch:

• A read-only download of model weights from Hugging Face (huggingface.co). This is a one-time fetch of approximately 3.7 GB. The request is one-directional inbound: Hush is downloading software (the AI program itself) from a software repository. Your prompts, conversations, or other personal data are not sent in this request — or at any other time. After this download completes, Hush makes zero network requests for the lifetime of the app.

Hugging Face’s privacy policy governs that single download: huggingface.co/privacy.

What Hush does not do

• No analytics. No tracking. No advertising identifiers.
• No third-party SDKs that collect data.
• No remote logging or crash reporting.
• No background activity.
• No cloud sync of conversations or settings.

Voice input and voice output

When you use the optional voice input feature, your speech is processed by an on-device speech recognizer. Hush uses one of two recognizers, both of which run entirely on your Apple Vision Pro:

• WhisperKit (preferred when available) — an open-source on-device speech-recognition library from Argmax Inc. The Whisper Base model used by WhisperKit is bundled inside the Hush app itself. The model file is part of the IPA you downloaded from the App Store; there is no separate download, no model server, and no network call at any point during speech recognition.
• Apple’s on-device speech recognition (fallback) — the system SFSpeechRecognizer running in its on-device mode, used as a fallback when WhisperKit isn’t available.

In both cases the audio and resulting transcript never leave your device.

When you use the optional voice output feature, Apple’s on-device text-to-speech synthesizes the audio locally.

Children’s privacy

Hush has no age-gated features and does not knowingly target users under 13. The app collects no personal information from anyone, including children.

Contact

Questions about this policy: juergen@klaassen.consulting.

Changes to this policy

If material changes are made to this policy, they will be published at this URL with an updated effective date. Hush does not have a mechanism to notify existing installs of policy changes.