Aim after you speak
Start dictating, then decide where it goes. Whisper-M captures the destination at the moment you finish, not the moment you started, so you can talk while you are still looking for the right field.
Whisper-M
Voice to text · macOS
音声入力
Speak anywhere on your Mac. The text arrives where you were aiming.
Press a global shortcut once and talk hands-free. Click into the field where the words belong, press again, and Whisper-M locks onto that destination, transcribes what you said, tidies it, and types it in. Transcription runs on your own machine through Core ML — with the network off, it still works.
Download for macOS
SHA-256
9d0353c941735735acc17df9b5f224102fa86ed3266352353f1841575f009345
Signed by Ludofax AB and notarized by Apple, so it opens without a security warning. If you would rather check that for yourself, run shasum -a 256 Whisper-M-0.2.2.dmg against the value above.
- Platform
- macOS 14+
- Transcription
- On device
- Account
- None
- Price
- Free
What it does
Dictation is the easy part. The awkward parts are aiming the text at the right place and reshaping it into something you would have typed — those are what most of this app is.
Runs on your Mac
Whisper models from Tiny English up to Large v3 run locally through Core ML. Apple Speech is available as an alternative where the device supports on-device recognition.
Modes you can edit
Raw, General, Prompt, Message, Note, Email and Code. The instructions behind each one are text you can read and rewrite — nothing is a hidden prompt you have to guess at.
Knows where it is
Routing can follow the app you are dictating into, so a terminal and a mail window do not get the same treatment. Dictionary replacements and voice snippets apply on top.
Never guesses at insertion
The destination is revalidated before anything is typed, and secure fields are a hard boundary the app will not write into. If insertion is not safe, the text goes to the clipboard instead.
Your history, locally
Dictations are kept on your Mac for 30 days by default, with usage summaries, immediate deletion, and JSON export and import that never carries your credentials.
Where your voice goes
Nowhere, by default. When a local model is selected, the dictation path makes no content-bearing network request — not as a policy, but because the code will not permit one. You can dictate with Wi-Fi switched off and confirm it for yourself.
Three things are worth being precise about, since “private” is a word every app uses:
- Downloading a model touches the network. Model files come from Argmax's Whisper repository on Hugging Face. That is a separate, explicit action — it downloads model weights and uploads nothing. Once a model is installed, that transcription path needs no network at all.
- Remote providers exist, and are off. You may point Whisper-M at an OpenAI-compatible endpoint — a model server on this Mac, a Mac mini on your desk, or a service you have chosen. If you do, your audio and text go there, and that service's retention terms apply, not ours. Plain HTTP is refused for anything that is not loopback.
- Recording only happens on a trigger. There is no always-on listening, no wake word, and no background meeting capture. Temporary audio is deleted after processing, and no transcript text, selected text or credential is ever written to a diagnostic log.
There is no account, no subscription, no analytics, no crash reporting and no sync. Nothing about you reaches Ludofax, including whether you installed the app at all. The full detail is in the privacy policy.
Installing
Ten minutes, most of it waiting for a model to download. Steps 4 and 5 are the ones people miss — macOS will not prompt for them, so they have to be granted by hand.
-
01
Open the disk image and drag Whisper-M to Applications
The image contains the app and a shortcut to your Applications folder. Drag one onto the other, then eject the image. Run it from /Applications rather than from the disk image — macOS ties permissions to where the app lives, and permissions granted to a copy on a mounted image are lost the moment you eject it.
-
02
Launch it and allow the microphone
Open Whisper-M from Applications or Spotlight. macOS asks for microphone access the first time it needs it; allow it. Without this nothing else in the app can work.
Whisper-M appears in the Dock while running and keeps a menu-bar item for starting and stopping dictation.
-
03
Choose a transcription model
In Settings → Transcription, choose Whisper models — on device and download Base Multilingual. It is the right starting point: accurate enough for everyday dictation and small enough to load quickly. It stays selected once ready; use the Use button when switching between downloaded models later.
Larger models are more accurate and slower. If you would rather not download anything, Apple Speech works instead — select it and grant Speech Recognition access when asked.
-
04
Grant Accessibility
Open System Settings → Privacy & Security → Accessibility and switch Whisper-M on. This is what lets the app type into the field you aimed at.
You can skip it. Dictation still works — the text goes to your clipboard and you paste it yourself.
-
05
Grant Input Monitoring
Open System Settings → Privacy & Security → Input Monitoring and switch Whisper-M on. This is what lets the global shortcut and mouse-button triggers work while another app is in front.
You can skip this one too, and start dictation from inside the app or the menu bar instead.
-
06
Switch back to Whisper-M
Return to the app after changing either setting. It rechecks both automatically and shows each permission's state separately, so you can see exactly what is still missing rather than guessing from a failure.
What each permission is for
| Permission | Needed | What it is used for |
|---|---|---|
| Microphone | Required | Recording your voice, only between an explicit start and finish. Without it the app cannot function at all. |
| Accessibility | Recommended | Reading which field is focused and typing the result into it. Withheld, every dictation falls back to the clipboard. |
| Input Monitoring | Recommended | Noticing your chosen shortcut or mouse button while another app is in front. Withheld, you trigger dictation from inside Whisper-M. |
| Speech Recognition | Only for Apple Speech | Using Apple's recogniser instead of a Whisper model. Never requested if you stay on local Whisper models. |
Whisper-M is deliberately not sandboxed. An app cannot watch for a global shortcut and type into another application's window from inside the App Store sandbox, and those two things are the product. That is also why it is distributed here rather than through the Mac App Store.
Requirements & honest limits
- macOS 14 or newer. Apple Silicon is the tested baseline. Nothing deliberately excludes Intel Macs, but they have not been tested and local model inference will be markedly slower.
- Disk space for models. The app itself is small; the model you choose is not. Base Multilingual is a modest download, Large v3 considerably larger.
- Version 0.2.2. This is early software that has been used daily by the person who wrote it and not yet by many other people. It is offered on that basis.
- Single Mac, single user. No login, no hosted history, no sync between machines. An iPhone companion and multi-device sync are separate future products, not features being withheld.
- Updates are manual. There is no automatic updater in this version. New releases are posted on this page.
Whisper-M is an independent project. It is not affiliated with Wispr AI, Inc.; “Wispr Flow” is named in our own notes only to describe the category of product this one belongs to.
Support
There is no crash reporting and no analytics in this app, which means nothing reaches us unless you send it. A short description of what you were doing just before something went wrong is genuinely the most useful thing you can include.
- Support
- info@ludofax.com
- Troubleshooting
- Common problems
- Privacy
- Privacy policy