Dictation That Stays on Your Machine: A Look at HEX
You hold a key combo, you talk, you let go, and the words show up in whatever app you're using. That's the pitch for most dictation tools—but a lot of them ship your audio off to a server somewhere. HEX takes that same simple interaction and does the transcription locally, on your own hardware, built in Rust. If you've ever wanted voice input without wondering where your voice went, this one's worth a look.
What It Does
HEX is local-first voice dictation for Apple silicon Macs running macOS 15 or newer. You hold a shortcut, speak, and release; HEX transcribes the audio on your Mac and pastes the result into the foreground app. During setup you download a local speech model, and ordinary dictation runs entirely on-device—no cloud transcription service required.
The app is built in Rust and lets you configure the basics: a shortcut, a language, and a model. It also has a feature called Modes, which handles corrections, application-specific behavior, and optional text processing. There are optional voice tools too—Voice Action and voice commands—both off by default, with provider-backed rewriting kept separate from the local speech recognition. The macOS app is available now, and there are Windows and Linux betas (x64 and ARM64 for Windows; x86_64 X11 and compatible wlroots Wayland compositors for Linux).
Why It's Cool
The local-first part is the whole point. Plenty of dictation apps are thin clients over a cloud API. HEX downloads a speech model during setup and runs it on your machine, so the everyday path—hold, speak, release, paste—doesn't touch a network. For anyone dictating sensitive material (notes, code comments, drafts you don't want stored elsewhere), that's a meaningful difference.
It's honest about what's optional. The README is careful to separate local speech recognition from the features that can reach out to a provider. Voice Action and voice commands start off. Provider-backed rewriting is its own thing. That kind of boundary-drawing is refreshing—you know exactly which parts stay on your device and which don't.
Modes give you per-app behavior. Instead of one global dictation setting, you can set up corrections, app-specific behavior, and optional text processing. If you want dictation to behave differently in your editor than in your chat app, that's the knob.
Installation respects your setup. The Homebrew cask installs the same signed app and keeps the built-in updater, and it explicitly won't migrate or forcibly replace an existing Hex.app. There's also a managed-deployment story: organizations can disable in-app updates (including manual checks) with a DisableUpdates preference in the com.kitlangton.hex2 domain, while model downloads keep working.
It tells you what to allow on your network. The README lists the hosts involved—downloads.hex.kitlangton.com and downloads.hex.kitlangton.dev for the app and updates, huggingface.co and its model CDN for local model setup, and download.moonshine.ai for optional Voice Commands. It even notes that a successful app download doesn't mean the model hosts are reachable—a detail that'll save you a confused half hour on a locked-down network.
How to Try It
The quickest path is Homebrew:
brew install --cask anomalyco/tap/hex
Then launch Hex from Applications and complete setup.
If you'd rather install manually:
- Download the latest signed DMG and open it.
- Drag
Hex.appinto Applications, then launch it. - Complete the permission and model steps in setup.
Setup will ask for a few permissions, each with a clear purpose: Microphone (so HEX can record while you dictate), Input Monitoring (so it can detect its shortcut globally—you'll drag the HEX tile into the list), Accessibility (so it can paste into the foreground app—same drag-into-the-list routine), and the transcription model download that runs locally.
HEX checks for signed updates automatically, and you can trigger a check with Hex > Check for Updates…. If you're coming from the original Swift app, the README notes you'll need to install the new app manually and complete setup again.
Grab it from the repository, and if you're on Windows or Linux, check the beta docs linked there.
Final Thoughts
HEX is a focused tool for a specific platform—Apple silicon Macs on macOS 15+, with Windows and Linux still in beta—so it isn't trying to be everything to everyone. What it does well is take the frictionless hold-to-talk interaction people like and keep the transcription local, with optional cloud features clearly walled off. If you dictate regularly and want your audio to stay on your machine, this is a straightforward thing to try. The fact that it's written in Rust and ships an SDK alongside the app suggests the project is thinking beyond a single binary, which is a good sign for where it goes next.