PilcrowHow it works
Three steps, no window.
Pilcrow lives in the menu bar. There is nothing to open, nothing to arrange next to what you were doing.
- 1
Tap the trigger key
Right Option by default. A quick tap on its own starts listening — holding it still works as a modifier, so the key keeps doing its normal job. Right Command and Right Shift work too, if your layout needs Option.
- 2
Speak
A small panel appears in the corner with a live transcript and an orb that reacts to your voice. It never takes keyboard focus, so the field you were typing in stays active.
- 3
Tap again
The text is pasted into whatever field had focus when you started — not whatever happens to be frontmost when you finish. If nothing editable is focused, it goes to the clipboard instead of being lost.
To: Dana Whitfield
Subject: Design review
Hi Dana,
Thanks for the notes — I'll fold the design review feedback in tonight.
Thanks for the notes — I'll fold the design review feedback in tonight.
The panel while you speak. It never takes keyboard focus, so the field you were typing in stays active — and the text is pasted, not typed, so it lands whole.
Any text field
There is no list of supported apps.
Pilcrow never asks which app you are in. It finds the text cursor and puts your words where it is — so the answer to “does it work in …” is yes, for anything you can type into.
- SlackMessage #generalOn it, deploying now
- MailSubjectThanks for the quick reply
- Terminal~ %git push origin main
- VS Code// explain the bug// retries on timeout
- NotesUntitledGrocery list for Sunday
- SafariSearch or enter website nameswift concurrency guide
- FigmaAdd a commentTighten this spacing
- MessagesiMessageRunning ten minutes late
- NotionType '/' for commandsDraft the launch checklist
- Xcode// TODO:// handle empty state
- RemindersNew ReminderPick up dry cleaning
- DiscordSend a messageAnyone up for a game?
- ChromeSearch Google or type a URLhow to notarize a mac app
- ObsidianStart typingIdeas from this morning
- PagesHeadingChapter one
- LinearIssue titleCaret jumps on paste
Some apps tell macOS nothing useful about what is focused — Slack, VS Code and Figma among them, along with most other Electron apps. Pilcrow pastes into them anyway. Refusing would have been the safer thing to build and the worse thing to use.
The trigger
One key you already have.
No chord to learn and nothing to reach for. Tap Right Option on its own and Pilcrow starts listening; hold it and it stays the modifier it has always been.
Prefer a different key? Right Command and Right Shift work too.
Privacy
Your voice never reaches a network.
Most dictation tools stream your audio to a server. Pilcrow uses Apple’s on-device SpeechAnalyzer, introduced in macOS 26, so the recognition happens on the same machine you are speaking to.
It asks this site whether a newer version exists, and Settings can switch that off. Nothing for transcription, analytics, or a licence check. Dictation works with Wi-Fi switched off.
Transcripts live in memory and are gone when Pilcrow quits. The recent-transcripts list in Settings is a ring buffer, never a file.
There is no sign-up, no e-mail field, and no identifier for you anywhere in the app. Nothing to leak, because nothing is collected.
The first time you dictate in a new language, macOS downloads that language’s speech model — a few hundred megabytes, once, from Apple. That is the only download the app ever causes, and it is handled by the system rather than by us.
The binding version of all this is the privacy policy.
The trade-off
Why most dictation apps still use a server.
Not carelessness — until recently it was the only way to get accuracy anyone would tolerate. The models good enough to run on a laptop only arrived with macOS 26. That is the whole reason Pilcrow can exist now and could not have two years ago.
Where a server genuinely wins
Server-side models can be far larger than anything that fits on a laptop, and they improve without you updating anything. Cloud tools also run on Windows, on Linux, and on Macs far older than macOS 26 — Pilcrow runs on none of those, and that is a real limitation rather than a positioning choice.
If you need the last few points of accuracy on difficult audio, or you are not on a current Mac, a cloud tool is the better answer and we would rather say so.
Details
The small things that decide whether you keep using it.
Right Option is the default trigger — tap it, don’t hold it. There is no Dock icon and no window to manage; Pilcrow starts at login if you want it to, and stays out of the way until you tap the key. Filler sounds are dropped from the transcript before it lands — and nothing else is touched.
Accurate. Lands in the right field.
Recoverable. Also left on your clipboard.
Invisible. Never steals focus.
Fluent. 30 languages, on-device.
What you said
Let’s see what happens. I mean, I am trying to add a lot of words. Uh. But I don’t know. I mean, uh, what can I say? Hmm. Yeah, I’m trying to say a lot of things. Let’s see.
What gets pasted
Let’s see what happens. I mean, I am trying to add a lot of words. But I don’t know. I mean what can I say? Yeah, I’m trying to say a lot of things. Let’s see.
Settings
Enough options, not endless ones.
Every setting has a default that works, so you can ignore all of them and still have a working dictation key.
Requirements
What you need
macOS 26 or later
The on-device speech recognition Pilcrow is built on does not exist in earlier versions.
Microphone access
So it can hear you. The audio is transcribed on your Mac and never leaves it.
Accessibility access
To watch for the trigger key, and to paste into other apps.
Both permissions take about a minute to grant and are covered in the help page.