Dictate anywhere on your Mac, without sending your voice to the cloud
Talking is about three times faster than typing. The catch has always been where your voice goes to become text.
Every dictation tool makes the same promise: hold a key, talk, and the words appear where your cursor is. The part they differ on, and the part that is hard to find on a pricing page, is whether your voice leaves your computer to make that happen.
That difference is not academic. If you dictate client notes, patient notes, anything under an NDA, or just a message you would rather not have stored on a server in another country, it is the whole decision.
The two ways a Mac turns speech into text
In the cloud. Your microphone audio is streamed to a server, a large model transcribes it, and the text comes back. This is how Wispr Flow works, and they have been straight about why: at the quality they want, they do not think on-device processing is practical at consumer scale yet. It is a real engineering position, and the cleanup you get from a big model is genuinely good.
On your Mac. A speech model - almost always OpenAI's Whisper, running locally - does the same job using your own CPU or Neural Engine. Nothing is uploaded. There is also nothing to meter, which is why local tools tend not to have word caps.
Some tools offer both and let you pick. Superwhisper does. Others are local by design, like MacWhisper, though that one is built for transcribing files you already recorded rather than for talking while you work.
What "anywhere" actually has to mean
A dictation tool is only useful if it works in the app you are already in. That sounds obvious and it is where most of them quietly fall down. The test is simple: does it type into every one of these?
- A Slack or Messages compose box
- A Gmail reply in a browser tab
- Your code editor
- A terminal, including one that is SSH'd into another machine
- A native app like Notes, Pages or Mail
The terminal case is the one that separates the tools. Anything that works by hooking into a specific app, or that only types into its own window and asks you to copy the result out, will fail at least one of these. A tool that pastes at the system cursor passes all of them, because as far as the Mac is concerned it is just a keystroke arriving.
This is also why Claude Code's built-in voice input is not a replacement for a dictation app, even for people who live in a terminal. It is scoped to that one CLI, needs a cloud round trip, and does not work over SSH. Different tool, different job.
The part nobody builds: doing something with what you said
Once the words are text, there is an obvious next step almost nothing takes. Sometimes you do not want your sentence typed out. You want it acted on.
Select an awkward paragraph in an email and say "make this warmer and shorter", and have it replaced in place, in the app you were already in. Point at a screen you do not understand and ask what is wrong with it. That is not dictation - the words are an instruction, not content - and it is the reason MightyMouse dictates into any Mac app and can act on what you said rather than only typing it out.
What local transcription costs you
An honest list, because the tradeoff is real:
- Heavy cleanup is weaker. A big cloud model is better at tidying long rambling speech. Local Whisper is accurate at getting the words right and does less rewriting on top.
- It uses your machine. Transcription takes about a second and some CPU. On a laptop running on battery you will notice it eventually.
- macOS only, in our case. Local models are built per platform. If you need the same dictation on Windows and on your phone, a cloud tool covers ground a Mac-only one does not.
What you get for that: no word cap, no per-minute charge, it works with no internet, and the question of who else has heard your audio never comes up.
Free local transcription, in every Mac app.
Nothing to meter, no word cap, no card to try it.
Common questions
Is local transcription accurate enough?
For getting the words right, yes - it is the same Whisper model family the cloud tools started from. Where a large cloud model still wins is heavy cleanup of long rambling speech. If you want that, hand the text to an AI backend afterwards, which is the same round trip a cloud tool does automatically, except you choose when.
Does it work with no internet?
Transcription does, because it runs on your Mac. Anything you then ask an AI model to do with the text needs a connection, the same as typing that question into its app yourself.
Does my audio get stored anywhere?
Not with local transcription - the audio is processed on your machine and never uploaded. Cloud dictation tools vary, and their privacy policies are the place to check.
If you want this on your own machine: MightyMouse dictates into any Mac app with free local transcription, and turns your iPhone into a trackpad for the same Mac.