Vox turns your Mac into a private AI dictation machine.
Hold Right Option, speak, release, and clean text lands wherever your cursor is. Change the hold key from the menu, dictate in English or French, and optionally translate the final output. Vox runs local Whisper on your Mac: no cloud transcription, no monthly subscription, no app-store platform risk.
==> Installing Hammerspoon, whisper.cpp, sox
==> Downloading local model
DONE. Hold Right Option and speak.
you:
Dictation should feel instant, private, and owned.
Most voice tools rent you access to a cloud workflow. Vox is a small local system you can inspect, modify, and keep.
Local by default
Audio is recorded and transcribed on the Mac using whisper.cpp. The default fast path does not send your voice to a cloud transcription service.
Fast enough to use all day
The Whisper server stays warm, the language is pinned, and deterministic corrections handle brand names without an extra LLM pass.
Operator vocabulary
Add client names, tools, jargon, and repeated corrections directly in the config. Vox gets better where your work is specific.
One-key workflow
Pick Right Option, Right Command, Left Option, or Left Command from the menu. Double-tap for hands-free lock. Release to paste anywhere.
Optional cleanup and translation
Ollama cleanup and translation can be toggled on when you want polished or translated output. Keep them off when raw speed matters most.
Open-source package
The download includes the Lua app, installer, sounds, README, and MIT license. Large models are downloaded during setup to keep the package light.
A little alien that learns you.
Vox started as dictation and grew a memory, a voice you can talk to, and a habit of fixing itself. Everything below ships in the free download — and everything stays on your Mac.
A memory that recalls by meaning
Every dictation is remembered locally. Ask for “the hospital thing” and it finds the note that never used those words — semantic recall with plain word-match as the always-on fallback.
“Hey Vox, …” — ask your own history
Start a dictation with “Hey Vox” and instead of pasting, the alien answers from its memory and your identity notes — on screen and in your clipboard.
“Hey Vox, remember…”
Save a fact by voice without pasting anything anywhere. A chime confirms it landed in the brain, permanently searchable and feeding every future answer.
Smart replies from your screen
Triple-tap the hotkey and Vox reads the window in front of you — that email, that Slack thread — and drafts the reply in your voice, using what it knows about you.
Rough idea → finished content
Hold Shift while dictating and a mumbled outline comes back as polished copy. Your identity notes keep it sounding like you, not like a press release.
Spoken commands
Say “scratch that” to undo, “new paragraph” or “new line” for breaks. Filler words are filtered, repeated sentences collapsed, stray commas cleaned.
A knowledge graph you can watch grow
Names, brands, and topics become linked entities automatically. One click opens your brain as a browsable wiki — or as an interactive graph of everything you talk about.
Dictation history, one click back
An app ate your words? The menubar keeps your last ten dictations — click any of them to paste it again. Nothing you say gets lost.
A tiny alien companion
A small alien idles at the bottom of your screen — click him for hands-free dictation, or use his buttons to speak-to-content or absorb the screen into memory. He dances when he’s warm.
Backs itself up daily
The brain snapshots to your iCloud Drive every day (newest seven kept). Move or merge your whole memory to another Mac with one export file.
Heals itself
Vox self-tests its own pipeline, notices when the engine drifts slow, restarts it, and re-verifies — plus watchdogs for stuck keys and stuck states. Boring reliability, on purpose.
Runs on almost any Mac
Apple Silicon gets the large model, older Intels get lighter ones automatically — a 2012 MacBook Air runs it. Slow Macs can even borrow a fast Mac’s engine over your LAN.
He lives at the bottom of your screen.
A tiny alien with a real job: listen when you hold the key, think when you ask, celebrate when the words land. Click a state — this is what he actually does on your Mac.
Questions people actually ask.
Is my voice really never uploaded anywhere?
Correct. Recording, transcription, the AI features, and the memory all run as processes on your Mac — whisper.cpp and Ollama, both local. There is no Vox server. Unplug your internet and dictation keeps working. The code is open source, so you don’t have to take our word for it.
macOS says “unidentified developer.” Is it safe?
Yes — that warning just means the app isn’t signed with an Apple Developer certificate yet. Right-click → Open, one time. Every line of code is public on GitHub under an MIT license, and the installer checksums the AI model it downloads.
Why is it free? What’s the catch?
No catch and no locked features. We build AI agent systems for businesses — Vox is our proof of work, in public. If you love it and ever want the same kind of automation in your company, you know where we are.
Will it run on my old Mac?
Almost certainly. Apple Silicon gets the most accurate model; older Intel Macs automatically get lighter ones. Our oldest test machine is a 2012 MacBook Air — it dictates in about four seconds. Any Mac from the last decade does noticeably better.
What happens to everything it remembers?
It lives in plain files on your Mac (a readable journal plus a small database), backs up daily to your own iCloud Drive, and exports as one file you can move to another Mac. Memory is optional — one menu toggle turns it off, and cancelled recordings are deleted outright.
How accurate is it with names and jargon?
Out of the box it’s Whisper large — excellent. Then it gets better: you can add your clients, brands, and jargon to a personal vocabulary, and Vox also learns the unusual words you use repeatedly. Corrections are deterministic, so a name you’ve fixed stays fixed.
Every dictation fires a synapse. Energy branches, connections form, and the brain literally grows — the more you talk to Vox, the smarter its answers get. This is a live simulation of what happens on your Mac; one menubar click opens your real graph.
Get Vox for Mac.
Enter your email and you'll land on the download page with the 2-minute setup guide. You will also be tagged as a Vox lead inside the existing AutomateScale form pipeline.
Install path — no typing needed
That’s it — it installs everything itself, ~5 minutes. If macOS says “unidentified developer,” right-click it → Open → Open (one time).
macOS asks for Microphone and Accessibility for “Hammerspoon” (Vox’s engine). The installer detects the grant and finishes itself.
Hold Right Option, speak, release — your words paste into the active app in ~1.5s.
curl -fsSL https://automatescale.com/vox/install | bash
Who this is for
Founders, operators, writers, builders, legal/documentation-heavy users, and anyone who wants faster text input without uploading every spoken draft to a third-party transcription platform.
What it is not
It is not a polished SaaS product, mobile app, Windows app, or hosted transcription service. It is a practical local Mac system built to be hackable and fast.
Ship faster by speaking.
Vox is one small example of the AutomateScale operating style: build the tool, package it cleanly, and turn useful internal infrastructure into a public asset.