Dictation for macOS
Hold right Option, talk, let go. Speechfy hears you, takes out the ums and puts in the commas, and the finished sentence appears in whatever app your cursor was already in. No keys to paste, no model to choose, no window to switch to.
Right Option out of the box, or any key or chord you record yourself. Speechfy checks the one you pick against what macOS has already claimed and tells you before you find out the hard way.
A small pill appears near your cursor and shows the sound of your voice as a row of bars that rise and fall with it. It never takes focus, so the app you were typing in keeps its caret and stays frontmost.
The text is inserted where your cursor was — the same message box, the same document, the same terminal. Nothing to copy, nothing to paste, no window to come back from.
What you said went to Speechfy in the EU, out to a model for the seconds it took to write, and came back as the sentence above. The audio and the text made from it are not retained after the request, not read by a person, and not used to train anything.
Four registers — Off, Verbatim, Clean, Formal — and Speechfy picks between them by which app you are dictating into, so a terminal never gets punctuation it did not ask for and an email never keeps your ums. You choose the register per destination once, watching the real model rewrite one of your own sentences four ways on screen.
Listen in Portuguese, write in English — the pair of controls is called "Listen in" and "Write in", and that is the whole feature. Translation is harder than tidying, so Speechfy says so next to the switch rather than in a footnote.
A dictionary of two halves: names to expect, which nudge the recogniser before it decides, and corrections to always swap, which fire on every dictation from then on. They are two separate lists with two separate buttons, because a hint and a find-and-replace are not the same promise.
Snippets turn a short spoken trigger into the longer thing — an address, a signature, the paragraph you send twice a week. The trigger is set in monospace next to the expansion, so a list of them reads as what it is at a glance.
| Trigger | Expansion |
|---|---|
| my email | [email protected] |
| my linkedin | linkedin.com/in/you |
| my standup | Yesterday: Today: Blockers: |
---
id: 8F3C1A7E-2D5B-4C91-9F0A-6B7E2C4D8A13
created: 2026-03-04T09:12:44Z
---
# Pin a note to the top
Idea for the app: let people pin a note to
the top of the list, and maybe show the
pinned ones different.
A notes room inside Speechfy that writes plain Markdown into a folder you chose — point it at the one you already have. Speechfy never deletes a note, and if that folder is in iCloud Drive or Dropbox it already syncs, by your account rather than ours.
History keeps the text of every dictation that went into another app, searchable and grouped by day, so the sentence you sent on Tuesday is findable on Friday. Nothing dictated into a note is in it, and nothing said while an excluded app was in front is either.
Words, days used, and how long you spent speaking them, with six months of days as one square each. There is no streak, no goal and no "you're on track" — those exist to make a renewal feel earned, and there is no renewal here.
Where it runs
Speechfy is a Mac app: Swift and AppKit, with the transcription and the polish done by Speechfy’s service rather than on your laptop. Windows and Linux are in development — what is drawn for them below is the design, and there is nothing to install yet.
macOS
Right Option out of the box, or any key you record yourself. Hold it, say the sentence, let go: the pill draws your voice while you talk and the finished text is inserted at the caret — the same note, the same message box, the same terminal.
Available now. macOS 14 or later, Apple silicon and Intel.
Windows
Hold the key, talk, let go, and the text is inserted into the window that already had focus. It is the Mac app's one mechanic carried across, pill and all — and it is in development, so what is drawn here is the design rather than a build you can run.
In development. There is no Windows download.
Linux
Hold the key, talk, let go. Reading the key from the input device rather than from the window server is what makes one build work on both — a shortcut registered with X11 is invisible to Wayland, and the reverse. It is in development, so what is drawn here is the design rather than a build you can run.
In development. There is no Linux download.
Privacy
Speechfy talks to its own service in the EU, and the service talks to the models that hear you and write your sentence out. There is no key to paste and no provider to choose — keeping the quality at the top is our work rather than yours. Some of those models run outside the EU, which we would rather say plainly here than have you discover in the policy; nothing is stored there, and when the request is finished what you said is gone.
TLS 1.3 in transit · AES-256-GCM at rest · Argon2id for passwords
Pricing
What you dictate is transcribed and polished by Speechfy’s own service, so what a plan sells is words: how many of them you can say in a month.
Showing monthly prices. Pro costs less by the year, and the line beside it is worked out from the two numbers on this page rather than typed in.
€0 a month
2,000 words a month. It resets on the first, and nothing else about it changes.
The whole loop rather than a demonstration of it: hold the key, talk, and the finished sentence lands in the app your cursor was already in.
macOS 14 or later.
First 10 days free
€9 a month
200,000 words a month — more than a working day of talking, every working day.
The whole app, for one person: every register, every language, everything you have already said, and an allowance nobody has reached by talking.
Ten days of Pro, free, with no card. Cancel from your account; it runs to the end of the period you paid for.
Per seat
Priced on how many of you there are, invoiced once.
What a company needs and one person does not: the people, the paperwork, and one bill.
The details, and a form, on the pricing page.
Everything each plan includes, row by row — with the six questions people ask before they pay, and the form for an Enterprise quote.