VaulType — Private Voice-to-Text for macOS

Built for Privacy. Designed for Power.

Every feature runs locally on your Mac — no cloud, no compromises.

Zero-Network Architecture

All processing happens on your Mac. No internet required. No data transmitted.

Dual AI Pipeline

whisper.cpp for speech recognition + llama.cpp for text enhancement, running locally.

6 Processing Modes

Raw, Clean, Structure, Prompt, Code, and Custom modes for any workflow.

Universal Text Injection

Text appears at your cursor in any app — editors, terminals, browsers, everything.

Voice Commands

Launch apps, manage windows, and control your Mac with voice commands.

App-Aware Context

Automatically selects the right processing mode based on your active application.

90+ Languages

Automatic language detection with support for 90+ languages worldwide.

Metal GPU Acceleration

Leverages Apple Silicon's Metal GPU for fast, efficient on-device inference.

How It Works

A four-stage pipeline that turns your voice into polished text — entirely on your Mac.

Your Voice

Speak naturally

whisper.cpp

Speech Recognition

llama.cpp

Text Enhancement

Any App

Text at Cursor

Six Modes. One Voice.

From raw transcription to structured code comments — choose how your voice becomes text.

Voice Input

i went to the store yesterday and buyed some stuff it was like really expensive and i dont know if we should of done that

Processed Output

I went to the store yesterday and bought some things. It was really expensive, and I don't know if we should have done that.

Voice Input

this function takes a user id looks it up in the database checks if they have permission and returns their profile data with caching

Processed Output

/**
 * Retrieves a user profile by ID with caching.
 *
 * Looks up the user in the database, verifies
 * their permissions, and returns the profile
 * data. Results are cached for performance.
 *
 * @param userId - The unique user identifier
 * @returns The user profile object
 * @throws UnauthorizedError
 */

See It in Action

A clean, native interface that stays out of your way.

Why VaulType?

Privacy-first by architecture, not by policy.

Zero cloud, by architecture

There are no cloud features to opt into. Speech recognition and LLM processing run entirely on your Mac — the only network requests are optional model downloads.

Free and open source

GPL-3.0 licensed. No tiers, no subscriptions, no lifetime licenses. Read the code, audit it, or build it yourself.

Both AI engines built in

whisper.cpp and llama.cpp ship inside the app with Metal GPU acceleration. No separate installs, accounts, or local server setups.

Six processing modes

Raw, Clean, Structure, Prompt, Code, and Custom — with app-aware context that picks the right mode for the app you are dictating into.

Voice commands

Launch and switch apps, run Shortcuts, and define custom aliases by voice. The direct version adds window management, volume, and system controls.

Native menu bar app

Swift and SwiftUI, built for macOS 14+. Lightweight, hotkey-driven (including the fn/Globe key), and out of your way until you need it.

Get VaulType

Free, open source, and ready in seconds.

Recommended

Homebrew

One command. Auto-updates included.

brew install --cask vaultype/vaultype/vaultype

Direct Download

Download the latest .dmg from GitHub Releases.

Download .dmg

Mac App Store

Sandboxed version with App Store updates. Copies dictated text for ⌘V paste instead of typing at your cursor.*

View on App Store

Build from Source

Clone the repo and build with Xcode.

git clone https://github.com/vaultype/vaultype

cd vaultype

open VaulType.xcodeproj

* App Store sandbox rules don't allow apps to type into other apps, so the Mac App Store version delivers text via the clipboard (a prompt reminds you to press ⌘V) and limits voice commands to launching/switching apps, running Shortcuts, and custom aliases. The direct download and Homebrew versions type text at your cursor and include the full command set, global shortcut aliases, plugins, and battery-aware throttling. Speech recognition and LLM processing are identical — and 100% on-device — in both.

Frequently Asked Questions

Everything you need to know about VaulType.

What is VaulType?

VaulType is a privacy-first, macOS-native speech-to-text application. It uses local AI models (whisper.cpp and llama.cpp) to convert your voice into text directly on your device, with no cloud processing.

Is VaulType free?

Yes! VaulType is completely free and open source under the GPL-3.0 license. No subscriptions, no freemium tiers, no hidden costs.

Should I get the Mac App Store version or the direct download?

Both are free and share the same on-device AI pipeline. The direct download (DMG or Homebrew) types text at your cursor and includes the full voice-command set, global shortcut aliases, and plugins. The Mac App Store version is sandboxed per Apple's rules, so it copies dictated text to the clipboard for you to paste with ⌘V, and voice commands are limited to launching/switching apps, running Shortcuts, and custom aliases. Choose the direct download for the full experience, or the App Store version if you prefer App Store installs and updates.

Does it work offline?

Absolutely. VaulType is designed with a zero-network architecture. All speech recognition and text processing happens locally on your Mac. No internet connection is ever required.

What languages are supported?

VaulType supports 90+ languages through Whisper's multilingual models (such as small, medium, and large-v3). English-only models are optimized for speed if you only need English. Language detection is automatic — just speak and VaulType recognizes the language.

What are the system requirements?

macOS 14 (Sonoma) or later. Apple Silicon (M1/M2/M3/M4) recommended for GPU acceleration. Minimum 8 GB RAM, 16 GB recommended for optimal performance with larger models.

How does privacy work?

Your voice data never leaves your device. VaulType processes everything using on-device AI models — no cloud APIs, no data transmission, no telemetry. Audio recordings are processed in memory and discarded — only the resulting text is stored locally in your history.

What are the processing modes?

VaulType offers 6 modes: Raw (exact transcription), Clean (grammar/spelling fixes), Structure (organized bullet points), Prompt (AI prompt formatting), Code (code comments/docstrings), and Custom (your own templates).

Can I use VaulType with any app?

Yes. The direct download injects text directly at your cursor position in any application — text editors, terminals, browsers, IDEs, email clients, and more. The Mac App Store version copies the text to your clipboard and prompts you to press ⌘V, which also works in any app.

How do voice commands work?

VaulType recognizes special voice command prefixes to launch applications, manage windows, and control system functions. Say things like "Launch Safari" or "Minimize window" while VaulType is listening. In the Mac App Store version, commands are limited to launching/switching apps, running Shortcuts, and custom aliases (sandbox rules exclude window and system control).

How do I choose the right Whisper model?

Start with the "small" model for the best balance of speed and accuracy. Use "tiny" or "base" for faster results on shorter audio, or "large-v3" for maximum accuracy on complex audio. Larger models need more RAM.

Your voice stays private. Text appears at your cursor.