Privacy Mask

Find personal information in text and mask it before you share it.

Made for the moment just before you paste an incident report, a log, or a customer record into Slack or a GitHub issue — when you realize there is a name in it.

Why this one

Everything runs on your Mac. Nothing is sent anywhere, and no rules are fetched from anywhere either.

It handles Japanese properly, which is the part other tools miss. Japanese personal names, addresses and My Numbers cannot be found by pattern matching, so this extension uses macOS's own detectors, Apple Intelligence's on-device model and, when it is installed, a name model trained for privmask instead:

What	Found by
Phone numbers, addresses	NSDataDetector — Apple's models, including full-width and unhyphenated Japanese formats
My Number	Pattern plus check-digit validation, so an order number is not mistaken for one
Email, postal codes	Patterns
Credentials — API keys, tokens, secrets	A published prefix, or the name that introduces the value
Your own terms	A term list you keep
Japanese personal names	Apple Intelligence, on device, and the privmask name model when installed
English personal names	NLTagger

How it works

Run Mask Text with something selected, or with text on the clipboard.

Everything found appears immediately, and the personal names follow a moment later. The model reads the whole text a chunk at a time, so a long passage takes longer than a short one, and there is no reason to make you wait for the rest. If a chunk cannot be read, the extension says how much of the text went unexamined rather than letting it pass as clean. Each finding shows how confident the detection is, in red, orange or yellow, and which detector found it. Everything is selected by default; press ⌘T to leave one alone.

↵ copies the masked text. ⌘⇧↵ pastes it straight into the app you came from.

Values are replaced with numbered placeholders — [NAME_1], [EMAIL_2] — and the same value always gets the same number, so a reader can still follow who is who. This is not reversible. No mapping is kept anywhere: that table would be a second copy of exactly what the masking removed.

Your own terms

Customer names, company names, project code names — the things no general detector can know are sensitive. One per line:

# ~/.config/privmask/terms.txt
株式会社サンプル商事
Project Bluebird

Matching ignores case and character width. Spelling variants are matched by the on-device model, so when that is unavailable the extension says so rather than letting you assume the list was fully applied.

Requirements

macOS 13 or later. Japanese personal names additionally need macOS 26 with Apple Intelligence enabled, or the privmask name model — pattern matching cannot find them.

The model runs by default wherever it is available. Where it is not, the extension tells you plainly that names were not looked for, rather than returning a clean-looking result: a gap you cannot see is worse than one you can. You can turn it off with the Use the on-device language model preference, which makes the extension much faster, at the cost of matching spelling variants of your own terms, and of Japanese names unless the name model is installed.

	13 – 25	26, Apple Intelligence off	26, Apple Intelligence on
Phone numbers, addresses	✅	✅	✅
Email, postal codes, credentials, My Number	✅	✅	✅
Your term list, matched exactly	✅	✅	✅
English personal names	✅	✅	✅
Japanese personal names	name model only	name model only	✅
Spelling variants of your terms	❌	❌	✅

The name model

A small model trained to find Japanese personal names, run with Core ML on any Mac this extension supports. It is tens of megabytes, so it is not bundled with the extension; it comes with the privmask CLI:

brew install snaka/tap/privmask

The extension finds it there without any setup. To use a copy somewhere else, set the Name Model preference to the directory holding ner.mlmodelc. Turn off Use the privmask name model when installed to leave it out.

Under the hood

Detection lives in privmask, a Swift package and CLI you can use on its own:

cat app.log | privmask

Developing

npm install
npx ray develop

The Swift side is fetched from the privmask repository, so nothing else needs to be checked out. Raycast runs a development extension from wherever its source lives, so keep the checkout in place while you use it.