Privacy Policy

Our Promise

EdgeWhisper is built on a simple principle: your voice data is yours alone. We don't collect it, we don't process it on remote servers, and we have no way to access it. This isn't a policy choice — it's an architectural guarantee.

Data Processing

All audio capture and transcription happen entirely on your Mac using Apple Silicon's Neural Engine via the MLX framework. EdgeWhisper runs the Voxtral Mini 4B Realtime model from Mistral AI locally — an open-source model released under the Apache 2.0 license. No audio data is ever transmitted to any server, cloud service, or third party. EdgeWhisper works fully offline after the initial model download from Hugging Face.

Data Collection

We collect nothing:

• No audio recordings
• No transcription text
• No usage analytics or telemetry
• No personal information
• No device identifiers
• No cookies or tracking pixels

Local Storage

EdgeWhisper stores the following data locally on your Mac in the Application Support directory:

• Your preferences and settings
• Personal dictionary entries
• Custom snippets
• Transcription history (for your review only)
• The Voxtral Mini 4B AI model files (~3 GB)

You can export or delete all local data at any time through the app settings. Uninstalling EdgeWhisper removes all stored data.

Permissions

EdgeWhisper requires two macOS permissions:

• Microphone Access: To capture your voice for on-device transcription. Audio is processed in memory and never saved to disk unless you enable transcription history.
• Accessibility Permission: To inject transcribed text into other applications via the macOS Accessibility API. This is required for universal app support.

App Store Subscriptions

Subscriptions are managed entirely through Apple's App Store. Payment information is handled by Apple — EdgeWhisper never sees or stores your payment details.

AI Model

EdgeWhisper uses the Voxtral Mini 4B Realtime model (~3 GB on disk), downloaded once from Hugging Face. The model runs entirely on your Mac — no cloud inference, no API calls. The model is open-source (Apache 2.0) by Mistral AI, ensuring full transparency and auditability.

Third-Party Services

EdgeWhisper uses zero third-party analytics, crash reporters, or ad networks. The only network request is the initial Voxtral model download from Hugging Face.

Children's Privacy

EdgeWhisper is not directed at children under 13. We do not knowingly collect any personal information from children.

Contact

EdgeWhisper is developed by Elitizon Ltd, founded by Raphaël MANSUY. If you have questions about this privacy policy, please contact us at contact@elitizon.com.