Free Mac app · nothing uploaded

Stop typing.
Just talk.

Hold one key, say what you were about to type, and it lands where your cursor was.

NEW MESSAGE
TOpriya@northwind.co
SUBJECTDeck review

Thanks for the quick turnaround — I’ll get you the revised deck by Thursday.

SENDDICTATING

How it works

One key. Three steps.

  1. 1Hold

    Hold the globe key. Or any chord you would rather use — it is a setting.

  2. 2Talk

    Say it the way it comes out. Any of the 99 languages Whisper knows.

  3. 3Let go

    The text lands at your cursor, in one piece. Nothing to copy, nothing to paste.

Record a call

It knows who said what. No bot in the meeting.

Your microphone is one track, whatever the Mac is playing is the other — so the transcript knows who said what without guessing at voices. Zoom, Meet, Teams, FaceTime, a phone on speaker: it captures the audio, not the app.

One microphone, several people — an interview, a room? Switch on Find speakers and it labels Speaker 1, Speaker 2. Rename one and the name follows through the transcript.

RECORDING01:52
YOU
OTHERS
STOP
  1. YOU 00:41

    Let’s keep the launch on the 14th, but ship the pricing page a week earlier.

  2. OTHERS 01:07

    Fine by me. Can you send the draft tonight so legal has time?

  3. YOU 01:22

    Tonight, yes. I’ll flag the annual discount — it’s the bit they’ll push back on.

  4. OTHERS 01:38

    Good. Then I’ll book the review for Thursday morning.

Your microphone is one track. Everything else is the other.

Then the notes

Notes that write themselves. Then you can ask.

Each recording gets a real title, the key points and who is doing what. Ask your notes a question and the answer cites the recording it came from, so you can open it and check.

Needs macOS 26 with Apple Intelligence switched on. That part runs on Apple’s on-device model, so it is as private as the rest.

PRICING CALL · 12 MARON THIS MAC

Pricing page before launch

KEY POINTS

  • Launch stays on the 14th; the pricing page ships a week earlier.
  • Legal needs the draft with time to review.

ACTION ITEMS

  • You — send the pricing draft tonight.
  • Priya — book the review for Thursday.
What did we decide about pricing?

00:41The pricing page ships a week before launch, with the annual discount kept.

READ 214 RECORDINGS · USED 1

Also

  • Drop in a file

    Any audio or video, 28 formats. Back comes a transcript with real timestamps, and every word follows the audio as it plays.

  • Search every word

    Everything you have ever said, findable from one box. Eight months ago, four seconds long — found.

  • Take it with you

    Export to a typeset PDF, Markdown or plain text. Your recordings stay yours, on your disk.

Nothing leaves your Mac

Switch the internet off. It doesn’t notice.

No account, no API key, no server. The speech models download once — about 730 MB — and after that everything is worked out on your own chip. Go on, flip the switches.

  • Wi-Fi
  • Bluetooth
  • Ethernet
  • The internet

honestly the best part is that none of this needs a connection, it just runs on the laptop and stays there

SENT0 BYTESCHECKED 21×
It didn’t notice.

The receipt

Every figure, checkable.

The price

Free

MIT licence. No account, no trial, no seat.

The app

14 MB

On disk. That is the whole download.

The models

730 MB

Two speech models, fetched once on first run and checked against a hash.

Sent

0 bytes

Per hour of speech. There is nowhere for it to go.

A call

2 tracks

Your microphone on one. The Mac’s own output on the other.

The interface

1 key

Hold. Talk. Let go.

Speed

377 ms

From letting go to text, median, under ten seconds of speech. Past ten it is slower — and the README says so.

Languages

99

Every one Whisper knows, including a sentence that switches halfway.

The source

Open

Read it. Run the benchmark on your own Mac.

Get VoiceDumps

Free. No account. Nothing uploaded.

117 DOWNLOADS SO FAR

LATEST  v1.2.2

For Macs with an Apple chip (M1 or later) on macOS 11 or newer. First launch fetches the speech models once, about 730 MB. Intel Macs and Windows are not supported yet. ALL RELEASES →

Questions

Does it work offline — and stay private?

After the first run, yes. The download itself is small because the speech models are not inside it — the app fetches them once, about 730 MB, and then never fetches anything again. From then on everything happens on your Mac: no account, no API key, no upload, and your audio and transcripts are never sent anywhere.

Which Macs can run it?

Apple Silicon Macs — M1 or later — on macOS 11 or newer. The current arm64 build does not run on Intel Macs. Windows is not supported yet.

Do the titles, overviews and ASK need anything special?

They run on Apple's on-device model, which needs macOS 26 with Apple Intelligence switched on. Without it — an older macOS, an ineligible Mac, or simply switched off — dictation, recording, transcription, search and export all work exactly the same, and ASK still finds and shows you the recordings that match. That half is on-device too, so nothing is uploaded either way.

How good is the speaker labelling?

Find speakers is off by default, and that is deliberate. On several clearly different voices it finds them all. On a two-person call it is well above chance but can split one person into two names — the surplus labels are sub-splits of the same speaker, so renaming still works. Meetings recorded as two tracks never need it: who spoke is already a fact there.

If macOS says it can’t verify the app

macOS says it cannot verify the app — is something wrong?

No. Nothing was scanned and nothing was found. The message means the app is not notarised, which needs Apple’s paid Developer ID at $99 a year; VoiceDumps is a free one-person project, so it is signed ad-hoc instead. macOS shows the same dialog for anything downloaded in a browser that it cannot trace back to a paid Apple account.

Do not click “Move to Bin”. It is the highlighted button, and it is the wrong one.

How do I open it anyway?

Either way is one-time. Installing from the terminal clears the flag that triggers the warning, so the dialog never appears at all:

curl -fsSL https://raw.githubusercontent.com/heynaavi/voiceDump/main/scripts/install.sh | bash

It downloads the latest release, checks its signature, copies the app across, asks for no sudo and writes nothing outside Applications. Read it first — it is short, and you should.

Already dragged the app to Applications? Clear the flag by hand, then open it normally:

xattr -dr com.apple.quarantine /Applications/VoiceDumps.app