Dictation for session notes that never leaves your Mac.

Most tools aimed at clinicians sell a medical vocabulary as the headline feature. Hushbird does not have one, on purpose.

Clinical terminology is treated exactly like everyday speech. The only vocabulary given special handling is the vocabulary you add: terms in your personal dictionary are corrected deterministically and passed to the cleanup model as known spellings. Everything else is left to context, and a guard rejects any rewrite that expands or drifts from what you actually said.

The result is that when it is unsure, you get something obviously wrong rather than something confidently wrong. On a clinical note, that is the failure mode you want.

Apple Dictation, Google Voice Typing and the cloud transcription tools all send audio to a server. Some will sign a BAA and some will not. Either way you are relying on a contract governing a copy of the session that already exists somewhere else, and a contract assigns liability after the fact rather than preventing anything.

Hushbird has no agreement to offer because there is nothing to sign an agreement about. No audio leaves, no transcript is stored anywhere but your Mac, no account exists and no server is contacted. That is something a compliance officer can verify in an afternoon rather than negotiate over weeks.

To be plain about it: compliance is a determination about your practice, not a property of software, and we are not the ones who make it. What changes here is that there is no third party for anyone to assess.

Double-tap the right Command key to start dictating and tap it once to stop. Double-tap the right Option key for the reader. Nothing is held down and there are no key combinations. It works from first launch with nothing to configure.

Speed is not the bottleneck. On an M4 Pro Mac mini a minute of speech transcribes in 0.48 seconds median across real use, measured by the developer. Cleanup adds well under a second. The constraint is how fast you can think and speak.

One tap of cleanup removes fillers and fixes grammar and punctuation. It corrects words misheard from context, and it is built not to rewrite what you meant or reorganise your thinking.

When you dictate, thoughts come out unstructured. That is the point of speaking rather than typing: you get the whole thought out before you lose it, in whatever order it arrived.

Reading your own writing silently does not work, because you read what you meant instead of what is on the page. Your eye slides over the sentence that doubled back on itself and the clause that never landed anywhere. Aloud, none of that survives. A run-on is obvious the moment you have to listen to it, and a word that is not quite the word you wanted sounds wrong even when it looks fine.

It is also how you check the cleanup itself. If tidying your grammar had changed your meaning, you would hear it immediately, where you would almost certainly have read straight past it.

Select any text, double-tap the right Option key, and it reads back in a natural voice. Tap once to stop.

Apple Silicon Macs only. No Intel Mac, no Windows, no iOS on the desktop app.

Dictation handles 25 European languages and detects which one you are speaking automatically. The reader covers fewer: English, Spanish, French, Italian and Portuguese.

Cloud models may still edge it out on very heavy accents or genuinely noisy rooms. The trial exists so you can find out on your own voice before paying.

It is a small indie app, not a funded team. There is no medical mode and no medical claims, because a general model substituting one real drug name for another is a risk we will not label as a feature.