Vox types what you say straight into whatever you are using — an email, a chat, a product description, a terminal. No window to switch to, nothing to copy, no waiting. A second and a half from releasing the key to finished text.
Interface animation: chat, email, terminal. Timings come from a real run on a MacBook Air M1 — about 1.5 s from releasing the key to inserted text.
There is no window to switch to first. Vox lives in the menu bar and types into whatever app you already have open.
🌐 fn or a shortcut of your own. While you hold it, Vox listens and shows a slim bar near the bottom of the screen.
In a normal voice, the way you would tell a person. The language is detected on its own, including two mixed in one sentence.
The text appears in the field — already punctuated, capitalised and stripped of “um” and “you know” by a second model.
Listing copy, review replies, supplier threads. Speaking a paragraph beats typing it.
Email, chat, support queues. A long reply out loud takes twenty seconds instead of two minutes.
Comments, commit messages, tickets, prompts. Works in the terminal and in the editor — anywhere there is a cursor.
RSI, wrist pain, a keyboard that fights back. Your voice does not get tired.
Measured on a MacBook Air M1. Transcription runs on Groq, the fastest inference I could find — and the difference is something you feel, not something you read about.
Your own vocabulary: product codes, brand names, jargon your team uses. No more “Kubernetes” coming back as “cuber netties”.
Raw text immediately, cleaned text after a beat, or raw first and quietly replaced with the clean version.
Tell it how you write: keep it formal, never use em dashes, always spell out numbers. Applied to every dictation.
The alternatives run $8–15 a month, forever. This is a one-time purchase, and it includes updates.
1,490 Telegram Stars (about $29) in @vox_dictation_bot. The key arrives in the chat the moment the payment clears. The bot speaks Russian for now — it is three taps; no Stars? Message me.
890 Telegram Stars (about $18). Signing up at Groq takes five minutes and an email address. Vox walks you through it with pictures.
| App | Price | Notes |
|---|---|---|
| Vox | $18–29 once | Cloud Whisper + cleanup, English & Russian, Intel too |
| Wispr Flow | $12–15 / month | Free tier: 2,000 words a week |
| Aqua Voice | $8–10 / month | Free tier: 1,000 words |
| Superwhisper | $8.49 / month or $249.99 lifetime | Free tier with local models |
| VoiceInk | $25 once | Local models, Apple Silicon only |
| macOS Dictation | Free | No filler-word cleanup or custom vocabulary |
Prices from each vendor’s pricing page, 17 Oct 2026. Full comparison.
macOS 15 Sequoia or later, Apple Silicon or Intel, an internet connection. The build is not notarised by Apple yet, so macOS asks you to confirm once on first launch — the exact buttons are below.
Drag Vox into Applications, the same as any other app.
Open Vox from Applications. macOS says it cannot verify the developer — click Done. Then System Settings → Privacy & Security → scroll to the line about Vox → Open Anyway and enter your Mac password. From then on it opens with a normal double-click. If macOS says the app is “damaged”, run the “Открыть Vox” (Open Vox) script from the disk image window: it clears the download quarantine flag.
The setup guide shows you where to click. Without the microphone Vox cannot hear you; without Accessibility it cannot type.
That is it. If macOS has claimed 🌐 for itself, Vox notices and offers ⌃⌥D instead.
The recording is sent to Groq for transcription (on the key-included plan it passes through our relay, which counts minutes and does not store audio). Vox does not record your screen and never watches what you type by hand. A short history of recent dictations stays on your Mac only, so you can recover text that did not land in a field — you can turn it off in Settings. The live preview while you speak uses macOS speech recognition. Your key lives in the macOS Keychain. Details in the privacy notes.
No. Transcription runs on a server, which is exactly why it is this fast and this accurate. Local models on a laptop are either several times slower or noticeably worse.
15 or newer. Apple Silicon and Intel both work.
Dictation pauses until the 1st. If that is not enough, plug in your own free Groq key and the limit disappears entirely — it is one setting.
No, to your email. Install it on your work Mac and your home Mac. The app checks the signature locally and never phones home to do it.
Not notarised yet. It is signed with my own certificate; the Apple Developer ID is on the list. That is why macOS asks you to confirm the first launch once — see Install above.
Built-in Dictation types as you talk and keeps the “um”s, repeats and false starts. Vox waits for the whole sentence, transcribes it with Whisper, has a second model fix punctuation and capitals and drop filler words, then inserts the finished text. Plus your own vocabulary and style rules. More in this comparison.
The first 7 days or 30 dictations are free and ask for nothing — no card, no account. Buy after that, if it earned it.