APPLE SILICON · MACOS 26+

Free, offline
dictation for Mac.

A little less typing. A lot more you.

VoxKey is a free, open-source dictation app built natively for Mac. Hold a key, speak, and turn your voice into text where you’re writing. Speech recognition and optional English cleanup run locally, so your audio and text stay on your Mac during processing.

Free and open source. Runs on your Mac.

English, Spanish and more · Version 0.7.1

HOW IT WORKS On your Mac
Explore the three steps
A note to the team

TUESDAY, 9:41 AM

Hey team, a quick thought.

Here’s what I have in mind…

Let’s keep the first version simple. We can try it together and build on what works.
Hold to speakListeningInserted
Focus a text field. Hold Globe/Fn to start.Say it naturally. VoxKey listens on your Mac.Release the key. Your words land in the text field.

An illustration of the flow. No microphone needed here.

BUILT FOR MAC

A native app.
Right where you write.

VoxKey lives in your menu bar, with native Mac settings and a keyboard-first workflow. Built with Swift and AppKit for Apple Silicon, it runs its speech and editing models inside the app. No local server or extra runtime to install.

OPTIONAL LOCAL CLEANUP

Fewer fillers.
Clearer English.

Turn on Improve transcription to remove fillers and repetitions and improve English grammar. Superwhisper’s S1-mini and a GECToR grammar model work together on your Mac, after their one-time downloads.

Off by default · English only · Up to 2.03 GB of models. The grammar model is for noncommercial use. Edits can change words; review important text.

How it works

100% LOCAL DICTATION

Your words stay
on your Mac.

Speech recognition runs on your device. No saved recordings, no dictation history. Works offline after model setup.

Model downloads and update checks use the internet. Your destination app and clipboard handle text after delivery or copy.

This website measures visits, clicks, and page loading times with PostHog using an anonymous browser identifier. No session recordings.

Privacy details

GET STARTED

Get started in three steps.

I built VoxKey as a free, open-source app.
Feedback and contributions are welcome.

Email hello@rodrigouroz.com

Install VoxKey

  1. Download the VoxKey DMG, open it, and drag VoxKey into Applications.
  2. Launch VoxKey, grant Microphone and Accessibility access, and choose a speech model during setup. Each card lists its languages; choose Download, then Use Model.
  3. Click a text field, hold Globe/Fn, and speak after the capture sound. Release to insert your text.

Questions

Can I dictate offline in English and Spanish?

Yes. Download a multilingual Whisper v3 Turbo model to dictate in English, Spanish, or other supported languages. Distil-Whisper v3 models support English only. After model setup, speech recognition works offline on your Mac. Model downloads and app update checks still use the internet.

What do I need?

An Apple Silicon Mac running macOS 26 or later. Choose an English-only Distil-Whisper v3 model or a multilingual Whisper v3 Turbo model for Spanish and other languages. You’ll need an internet connection for model downloads; you can keep several models and switch in Settings.

Will it work in every app?

Editor compatibility and real-world latency are still being tested. Try it in your usual text fields. If delivery can’t be confirmed, recover your text with Safety Net. If something goes wrong, report the app, macOS version, and VoxKey build, using made-up text.

How do updates work?

Choose Check for Updates in VoxKey’s menu. Updates wait until dictation and recovery finish, and ask before restarting. Downloads are also available on GitHub Releases.

Can it improve my transcription?

Basic dictation works on its own. Improve transcription is optional, defaults off, and removes fillers and repetitions while improving English grammar. Its models download separately (up to 2.03 GB) and run on your Mac. Improvement pauses for other languages and automatic detection. The grammar model is for noncommercial use and may change your words. Review the model terms and testing notes before enabling it.