Voicey

On-Device Processing

Voicey runs Qwen3 speech recognition models directly on your Mac using native Swift MLX and Metal on Apple Silicon. This means:

• Your voice audio never leaves your device
• Transcription happens entirely on your Mac
• No internet connection required for transcription
• Works offline after initial model download

Network Access

Voicey only uses network access for one purpose: downloading AI models. This happens:

• On first launch when you select a transcription model
• When you choose to download additional models
• Models are downloaded from Hugging Face's model repository

After models are downloaded, Voicey works completely offline. No audio, text, or usage data is ever transmitted.

Data Storage

Voicey stores minimal data locally on your Mac:

• AI Models — Downloaded Qwen3 ASR models for transcription
• Preferences — Your settings (hotkey, selected model, UI preferences, spelling and context options)
• Custom vocabulary — Terms you add for spelling steering (stored locally if you use that feature)

Voicey does not store your transcriptions. Text goes directly to your clipboard and is not saved by the app.

Spelling & Context (Optional)

If you enable spelling and context features, Voicey may read text on your Mac to improve recognition of names and jargon:

• Custom vocabulary — Terms you list are passed to the on-device speech model before transcription
• On-screen text — With Accessibility permission, Voicey can read the focused field and nearby UI text in the front app at the start of a recording
• On-device OCR — If you opt in and grant Screen Recording permission, Voicey can capture the target window and run Apple's Vision OCR when Accessibility exposes little text (for example in some Electron apps)

This text is used only locally to steer transcription. It is not sent to Voicey servers — we do not operate servers that receive your data.

Permissions

Voicey requests only the permissions needed for the features you use:

• Microphone — Required to capture audio for transcription
• Network — Used only for downloading AI models from Hugging Face
• Accessibility — Optional. Used if you enable auto-paste into other apps, or “Use text from the target app” for spelling steering. Text is read on your Mac only.
• Screen Recording — Optional and off by default. Used only if you enable on-device OCR for spelling when Accessibility does not expose enough content. Window capture and Vision OCR run on your Mac; nothing is sent to the cloud.

We follow the principle of least privilege — requesting only what's needed for the features you turn on.

No Telemetry

Voicey includes no analytics, crash reporting, or telemetry of any kind. We don't know:

• How often you use the app
• What you transcribe
• Which features you use
• If the app crashes

Your usage is completely private.

Open Source

Voicey is open source. You can review our code at any time to verify these privacy claims:

View Source Code

Contact

If you have questions about this privacy policy or how Voicey handles data, please open an issue on our GitHub repository.