How tabi works, in plain English. Anything missing, just ask.
A Mac app that starts as dictation and does not stop there. Hold a key, say what you mean, let go, and cleaned up text lands wherever your cursor was, in any Mac app. Say “hey tabi” and it answers or rewrites what you selected. Select any text and it reads it back.
Every other dictation app stops when the text lands. tabi keeps going: say “hey tabi” and it answers at your cursor, rewrites what you selected, or replies to the message you selected. Select any text and it reads it back in a real voice, or drop in a PDF or a book and it reads that too. A word it keeps getting wrong, you fix once and it remembers. And every word runs on Mistral, the French AI company, and nowhere else.
It turns a take into a request. Start with “hey tabi”, ask a question, and the answer lands at your cursor. With nowhere to type, on the desktop or a page with no box, it shows in a small box above the orb instead. With text selected, say what to change and the rewrite lands in its place, or say how to reply and tabi puts the reply on your clipboard and pastes it where you click next.
Hold fn and talk, then let go. Tap fn twice for a hands free take that keeps going until you press fn once more.
Yes. tabi sits in the Dock like any app, and closing its window leaves it there, ready. If you would rather it lived only in the menu bar, switch on "Hide tabi from the Dock" under Settings, General. The Dock icon then shows only while tabi's window is open.
Yes, with one catch. A keyboard that is not Apple's handles its fn key inside the keyboard, so the Mac never hears it and tabi cannot either. Pick a second hold key in Settings, right Option or right Command. It works beside fn, so the built-in keyboard keeps fn.
Yes. Each app can own a style, so what you say in Slack is not cleaned up the same way as what you say in Mail.
Thirteen, and it hears the one you switch into: English, Français, Deutsch, Español, Italiano, Nederlands, Português, Русский, हिन्दी, العربية, 中文, 日本語 and 한국어. You can let tabi detect the language or pin a single one in Settings.
tabi is an independent app made by Nikhil Bhima. Every feature runs on Mistral AI's models. tabi is also part of Anthropic's Claude for Startups program. Claude helps with the engineering work and never runs inside the app. Read the announcement.
Yes. Select text in any app and press ⌃⌘R, or pick Read my selection aloud from tabi’s menu bar icon. It reads in a real voice, and the first words come within about half a second.
Yes. Open Listen in tabi and choose a file, or drop one on the window: a PDF, a Word file, an EPUB, a text file or a photo of a page. tabi shows a PDF as its own pages and lays out any other file cleanly, leaving the page numbers and running headers unread, and reads it aloud with the sentence tinted and each word marked as it is said. Word, EPUB and text files, and PDFs that hold their own words, are opened on your Mac; scanned PDFs and photos are read by Mistral OCR.
Yes. In Listen, pick a language under Read it in and tabi translates the text and reads it in that language, starting within a couple of seconds. It speaks nine: English, French, Spanish, Portuguese, Italian, Dutch, German, Hindi and Arabic. A switch shows the text as written next to what you hear.
Yes. In Listen, press the save button in the player and tabi saves the whole text as an M4A file, in the voice and language you picked, to play on your phone or anywhere else.
Yes. Select it and say “hey tabi” with the change you want, like “hey tabi, make this friendlier”, and the rewrite lands in its place.
Yes. In the Dictionary, add the right word and what it wrote instead. From then on it writes what you meant. Names and terms it should know ahead of time go in the Dictionary too. Or correct it where it landed, and tabi learns the fix.
To Mistral, the French AI company. On the way they pass through tabi's relay, a small program on Cloudflare that forwards each call and keeps none of it. No other AI company sees a word.
No. Claude helps build tabi; it does not run it. Everything you say or write in tabi goes to Mistral and nowhere else.
No. The audio is discarded the moment it becomes text. Until then it sits in one private file on your Mac, so a dropped connection cannot lose what you said, and that file goes the moment the take lands. Mistral, which turns your speech into text, keeps what it processed for thirty rolling days to catch abuse, then deletes it. Your takes are saved as text on your own Mac, and you can delete any one of them, or all of them, inside the app.
No accounts, and nothing that follows you between sites. The app collects no analytics at all; the only things that leave it on their own are an update check and a crash report with no words in it, and Settings has a switch for each. This site counts page views in aggregate, with no cookies and nothing tied to a person, so there is no consent banner to click.
Yes, while tabi is in beta, with every feature switched on: dictation, read aloud, Listen, hey tabi, every voice and a clone of your own. There is nothing to sign up for: press Download, open the app, and it signs itself in. If that ever changes, the app tells you first.
Yes. Speech to text and cleanup both run on Mistral's models, so a take needs a connection.
A Mac running macOS 14 or later.
If you let tabi switch off what a press of the globe key does, macOS keeps it that way after tabi is gone. Open System Settings, then Keyboard, and set “Press globe key to” back to what you like, such as Show Emoji & Symbols. If you turned off Apple’s Dictation shortcut for tabi, it stays off too: in the same Keyboard settings, under Dictation, pick a Shortcut again.
Still stuck? Write to hello@heytabi.app. A real person reads it.