Privacy policy

Effective date: November 7, 2026

This policy describes the Whisper Speech macOS app, version 1.0.0.

Local processing

Whisper Speech transcribes audio on your Mac using bundled whisper.cpp software and downloaded speech-recognition models. It does not send your recordings, imported audio, transcripts, or saved prompts to the developer or to a transcription service. The app has no user account and includes no analytics, advertising, or tracking SDKs.

Audio and permissions

Microphone access lets you record your voice. Importing a file lets the app read and convert the audio you select. Imported original files are not copied into history or deleted by the app.

Meeting recording uses your microphone and the selected application's audio. For a browser, this includes audio from all of its tabs, not just the meeting tab. Meeting recording requires your action and the relevant macOS microphone and screen/system-audio permissions. ScreenCaptureKit supplies audio; the app does not save screen video frames.

Meeting suggestions inspect Core Audio application-activity metadata to identify apps using the microphone. A suggestion does not start recording. You can turn suggestions off in Settings.

Input Monitoring supports the optional global recording hotkey. Accessibility permission supports optional automatic pasting into another application. You can use the recording button and copy text manually without those optional permissions. Revoke permissions at any time in macOS System Settings > Privacy & Security.

Model downloads

Speech-recognition models are downloaded from Hugging Face after you choose to download them. The model offered by default is approximately 0.5 GB; other available models have their sizes shown in the app. These connections download model files, not upload your audio or text. Hugging Face and its download infrastructure receive normal connection information, including your IP address and the requested model, under their own privacy policies. Once a model is installed, transcription works without an internet connection.

Local storage and retention

Models and history are stored in the app's local application-support directory, inside its container for sandboxed App Store builds. Interface preferences, recognition settings, the history limit, and saved prompts are stored locally using macOS UserDefaults.

By default, history retains up to 10,000 items. A stopped recording is saved as a WAV audio file before transcription, and a successful nonempty transcript is saved as a TXT file. Saved audio can remain when transcription is cancelled, fails, finds no speech, or the recording is too short. Audio-only items count toward the history limit even though they are not shown as transcripts in the History list. Oldest items are deleted when the limit is exceeded.

Temporary audio and processing files are normally removed after processing or cancellation. A crash or an error saving or combining audio can leave temporary WAV or CAF files on your Mac. When an error preserves audio, the app displays its location. There is no automatic recovery of an unfinished recording.

The app does not encrypt history files separately from the protections provided by macOS. Files you export, operating-system backups, and copies you place in synchronized folders are managed by you and the relevant system or service.

Your controls

Delete a transcript in History to delete its saved TXT and matching WAV. Settings > Disk usage > Delete saved audio files removes saved WAV files, including audio-only items, while keeping transcript text. Settings > History lets you change the retention limit. Applying a limit of 0 deletes saved history, including its audio, and disables future history saving. This does not delete files you exported, imported originals, saved prompts, or models.

You can delete downloaded models in Models and edit or delete saved prompts in the prompt picker. You can cancel recording or transcription, turn meeting suggestions off, and revoke macOS permissions. Copies in backups or other applications require deletion through those systems.

Copying, pasting, and exporting

Copy and hotkey dictation place transcript text on the macOS clipboard, replacing its previous contents. Copy with prompt combines the selected prompt and transcript on the clipboard; it does not call an AI service or generate a summary. Automatic pasting sends the text to the original application when permitted. The clipboard, clipboard-management software, and any application or service into which you paste or export data can handle that content under their own settings and policies. You control those destinations.

Policy availability

This policy is bundled with the app and is available offline in Settings > Privacy Policy. Changes to the app's data practices will be reflected in an updated policy distributed with the app.

Contact

For questions about this app or policy, email contact@oddly-interesting.org.