For Mac
Smart dictation that never leaves your Mac.
Hold a key, speak, and watch your words appear in any app. And the ums, stutters, and second thoughts? Cleaned up on the way in. No internet needed.
Free to use. Apple silicon Mac, macOS 14.4 or later.
A demonstration. While the talk key is held, a card hangs from the top of the screen showing the words as they form, including one the engine corrects and marks. The person says a filler, changes their mind about a time, asks for a new paragraph, and speaks an email address. When the key is let go, the cleaned text appears in the document at the cursor: "Meet me at ten." then a new paragraph, "Email team@example.com."
How it works
Hold a key. Talk. Let go.
That is the whole thing. It works the same in Mail, Slack, Notes, a browser, or a code editor.
Hold the talk key
Right Option, out of the box. Pick another key if you want, or set it to tap once to start and tap again to stop.
Your Mac turns it into text
The speech model runs on the Neural Engine. A short sentence comes back in well under a second.
The text is typed at your cursor
Cleaned up, and fitted to the sentence you were writing. Your clipboard is put back the way you left it.
The cleanup
What Slate fixes before it types.
Each one is a fixed rule. Your wording is never rewritten by an AI.
“Um, so the deadline, the deadline moved to Friday.”
So the deadline moved to Friday.
“I’ll see you at 10:30. Actually, I’ll see you at 11.”
I’ll see you at 11.
“Ship it Tuesday. Scratch that. Ship it Wednesday.”
Ship it Wednesday.
“Email sam at example dot com.”
Email sam@example.com.
All of it can be switched off in Settings.
Inside the app
What else it does.
Removes ums, uhs, and repeats
Filler sounds go. “The the” becomes “the”. A phrase you started twice is written once.
Keeps your correction, not your first try
Change your mind mid-sentence and only the second version is written. Say “scratch that” and the sentence before it goes.
Say “new paragraph” and get one
“New line” starts a line. “Open bracket” and “open quote” wrap what comes next. “Scratch that” takes back the sentence before it.
A dictionary for names and jargon
Add what you say on one side and what should be written on the other. Islamic phrases are spelled the same way every time.
25 languages, all on this Mac
The English model is the most accurate. The multilingual model covers 25 European languages and works out which one you are speaking.
Leaves your clipboard and music alone
Whatever you copied comes back, images included. Music pauses while you talk and starts again after. Sitting idle, Slate uses 0% CPU.
The promise
Nothing is sent anywhere.
Cloud dictation apps send a recording of your voice to a server, charge every month, and stop working on a plane. Slate does the whole job here.
- Speech becomes text on this Mac, on the Neural Engine, using the Parakeet model. There is no server to send it to.
- No account, no sign-in, no tracking. Slate does not collect anything.
- Zero network connections while dictating. Measured with lsof during a live transcription: none.
- Works with no internet after the speech model downloads, about 600 MB. Switching to the multilingual model fetches that one too.
- The rest of the list, in full. A license key is checked with the store once. Update checks, if you switch them on, ask this site for one small file and download the new version if you install it. Nothing about you goes with any of it.
# nothing. keep talking, run it again, still nothing.
$
Do not take our word for it. Run Little Snitch, LuLu, or Wireshark and watch while you dictate. Or turn off Wi-Fi and keep working.
The difference
Slate and cloud dictation.
| Slate | Cloud dictation apps | |
|---|---|---|
| Where speech becomes text | On your Mac | On their servers |
| How long it takes | Under a second | A round trip to their server |
| On a plane | Works | Does not work |
| Your voice | Never leaves the Mac | Uploaded |
| Account | None | Required |
| Price | Free | $144 a year and up |
Get Slate.
No account to make, and nothing to sign up for.
Download Slate for MacFirst launch asks for two permissions, and downloads the speech model.
Microphone
Used while you hold the talk key, and at no other time.
Accessibility
Notices the talk key, and puts the text at your cursor. Every other keystroke passes through untouched.
The speech model
Downloads once, about 600 MB. After that, dictation needs no internet.
FAQ