HARKNOTE

On-device. Recording, transcription, speaker ID, summaries and Ask all run locally. Nothing is sent off your device for processing.

No account. No sign-up, no profile, no cloud login. It just works.

No tracking. No analytics, no telemetry, no ads, no tracking SDKs.

Works offline. Once model assets are installed, recording and transcription do not need a cloud service.

Your choice of engine. NVIDIA Parakeet v3 or Whisper on both platforms, and you can transcribe a finished recording again with a different model. On iPhone, Parakeet transcribes 25 European languages and works out which one it’s hearing on its own; Whisper’s Large model is multilingual, its Base and Small models are English-only. On Android, Parakeet is the recommended default and covers the seven languages the app ships in; Whisper Tiny and Base are English-only. The app itself speaks English, German, Spanish, French, Italian, Dutch and Portuguese on both.

Summaries and Ask. A title, key points, decisions and action items after every session, plus a chat about what was said. Apple’s on-device model on iPhones with Apple Intelligence; Gemma 4 or Qwen3.5 through LiteRT-LM on Android phones that can run them; a built-in deterministic engine everywhere else.

Apple Watch and Wear OS. Record on your wrist with no phone nearby and no internet. Sessions transfer to your phone when it’s back in range and are transcribed there. Audio goes to your own phone, never to us.

Meetings, imports, exports. Optional read-only calendar access links a recording to its meeting. Import audio from other apps through the share sheet. Share a transcript as text or SRT captions when you choose to.

Yours to keep. On iPhone, private Apple iCloud storage can keep your app database available on your own devices, and audio sync is your choice. On Android, everything stays in the app’s private storage. Delete sessions anytime.