Note taker

Where the recording goes

Nowhere. The audio is recorded into this tab's memory, and the speech models run on this device's own processor. The only things downloaded are the models themselves, once, from Hugging Face; nothing is sent back. So there is no copy on a server, no account holding your history, and nothing for anyone to leak, subpoena or train on.

Privacy mode

On for a clinical visit, and a switch for everything else. While it is on:

• nothing is saved in the browser, so closing the tab ends the session for good;
• the session wipes itself after 15 minutes without use, with a warning first;
• this page stops sending even anonymous usage statistics;
• "Wipe everything now" is always one click away.

For clinicians

Because the recording never leaves your device, we never create, receive, store or transmit your patients' health information. The rest is yours, as with any tool: an encrypted device with a screen lock, and care over where the exported note goes. The transcript is a draft for you to check, not a medical record, and it is not for diagnosis. Several US states require everyone's consent to record a conversation; the consent box is there so asking becomes a habit.

Languages

Pick the language before you start. If the conversation moves between two, pick both: while recording, a button for each appears, and a tap marks the switch. Each stretch is then transcribed in the language it was in. You can also have everything written up in English, whatever was spoken.

Automatic detection is there, but it guesses from the first few seconds and can guess wrong on a short or quiet clip. Choosing the language is always more accurate.

Getting the best transcript

• Put the device between the people talking, not in a pocket or a bag.
• Tell it how many people are speaking if you know; it is then never wrong about that.
• Use Accurate for noisy rooms, strong accents and anything not in English.
• Click any time in the transcript to hear that moment, and correct words by typing over them.