Free · Open source · Runs on your Mac
Dictation that stays on your Mac.
Hold a key, talk, and the words land in whatever app you're in. By default your Mac does the transcribing and no audio leaves it. No account, no subscription.
Apple Silicon (M1 or later) · macOS 26 or 27 · Release notes
Macalla is for Mac for now. Windows and Linux builds follow. Release notes
- macOS· Apple Silicon (M1 or later) · macOS 26 or 27
- Windows· coming soon
- Linux· coming soon
Hey Sam, moving our sync to Thursday at ten. I'll send the revised deck tonight.
Three steps.
- 01
Hold a key
One shortcut works in every app, and you choose the key. Let go to stop. For a longer dictation, start it from the menu bar icon instead.
- 02
Talk normally
Your Mac transcribes the speech, then adds punctuation and capitals and takes out the fillers.
- 03
Text appears
It is pasted at your cursor: in Slack, your editor, the browser, wherever you were typing.
Your words, not ours.
Speech engines get your vocabulary wrong, and nothing else fixes it. Macalla keeps a dictionary of the words you actually say, built from your own corrections, and applies it on top of whichever engine is doing the listening.
- Every mis-heard word in your history is one click from becoming an entry, prefilled with what the engine heard.
- The dictionary sits above the engine, so it survives a change of engine, platform or model.
- Export it as JSON, and import that, a plain word list or heard-and-written pairs, so a team can share a vocabulary and you can carry yours between Macs.
| Engine, unassisted | Correct |
|---|---|
| Apple SpeechTranscriber | 6 of 15 |
| Parakeet TDT v3 | 9 of 15 |
Neither produced kubectl, pgvector, Supabase, Ogham. Measured on Apple M1 Pro, 16 GB, macOS 26.6.2.
Yours, by default.
Many dictation apps charge every month and transcribe on their own servers. Macalla does the job on your Mac, and by default nothing you say leaves it.
| Feature | Macalla | Subscription apps |
|---|---|---|
| Where audio is processed | Your Mac | Usually their servers |
| Price | No cost | Often a capped tier, then a monthly plan |
| Works offline | Yes | Depends on the app |
| Source code | Open (MIT) | Usually closed |
| Your vocabulary | Yours, on every engine | Usually tied to their app |
Local first. Cloud only when you ask.
Nothing leaves your Mac unless you choose cloud transcription, or Enhancement that runs anywhere but this Mac. Audio is transcribed on your Mac and never saved; the text stays in History on this Mac, 30 days at most, or less if you change History's setting.
- Runs on your Mac
Transcription, punctuation and filler removal happen on your Mac by default. Enhancement that runs anywhere but this Mac, on your own server or a cloud model, is sent your words at most once, when they are ready, and never while you speak.
- No account
Nothing to sign up for and nothing to log in to.
- Nothing sent home
Nothing about your use goes anywhere. No telemetry, and no account for it to be attached to.
- Readable source
MIT licensed. Read it, audit it, or build it yourself.
Questions
What happens to my audio?
The audio isn't kept. Your voice is turned into text on your Mac and the audio is discarded; it is never written to disk. The text is kept in History on this Mac for 30 days at most, and you can shorten that or switch history off. With a cloud engine turned on, the audio goes to that provider instead.
Does it work in every app?
In any app that accepts a paste. The text is pasted at the cursor, and where paste fails it is typed character by character.
Does it work without internet?
Yes, once your language's speech data is on the Mac. macOS downloads it from Apple once, then it runs on your Mac.
What languages?
Apple SpeechTranscriber, the engine built into macOS, reports 10: English, German, Spanish, French, Italian, Portuguese, Chinese, Cantonese, Japanese, Korean. We have measured English, German, French, Spanish and Italian. The last three were measured on the downloadable models, with synthetic voices. Other languages are not tested yet. The accuracy figures on this page are for English only, and the app's own interface is English.
Can I use my own API key?
Yes. Cloud transcription is off by default; you add your own key, it stays in your keychain, and the overlay names the provider while it listens. Nothing will ever be billed through Macalla.
Can it transcribe a recording?
Yes. Open a recording on the Files page and it is transcribed, or open a transcript you already have; either can then be summarised. The command line does the same. WAV and AIFF files need nothing else. On a Mac, mp3, m4a, mp4, mov, flac and opus work too; WebM and anything else need ffmpeg.
What hardware do I need?
A Mac with Apple Silicon (M1 or later), running macOS 26 or 27, the two versions it has been tested on. Every number on this page was measured on Apple M1 Pro, 16 GB, macOS 26.6.2, and nothing is claimed for other hardware.
Is it really free?
Yes. MIT licence, no paid tier, nothing held back for one. If it saves you a subscription or just helps you get more done, star the repository.
What happens to my transcripts if I stop using it?
They are in one file on this Mac and nowhere else. Delete it and they are gone; a deleted entry is overwritten in the file, not just hidden.
Windows and Linux?
Windows and Linux are next. The app is built cross-platform from the start, so the wait is for the builds, not a rewrite.
Macalla is a weird name!
Macalla is Irish for echo, literally "son of a cliff". Scottish Gaelic has mac-talla, "son of the rock". Both carry the idea of a voice sent back from somewhere, and we liked the feel of that. Say it "mah-KAL-uh", with the stress in the middle.
Ogham remembers. Macalla listens.
Macalla comes from the same people as Ogham, a shared memory for your tools. Dictation works on its own. Sending your transcripts into Ogham will be optional, and off unless you turn it on.
About Ogham →