JustSpeak

Built with Claude Code - AI-assisted development from concept to completion. 🤖

A lightweight macOS menu bar app for offline voice dictation, with optional cloud LLM post-processing. Hold Command, speak, and your words appear at your cursor.

Features

• Offline Transcription - Core speech-to-text runs locally with Whisper.cpp
• System-Wide - Works in any application where you can type
• Simple Hotkey - Hold Command to record, release to transcribe and paste
• Cancel Anytime - Press Escape while holding to cancel recording
• Optional LLM Cleanup - Route transcribed text to an OpenAI-compatible endpoint before paste
• Privacy Control - Keep everything local or opt in to cloud post-processing
• Accurate - Powered by OpenAI's Whisper large-v3-turbo model
• Native macOS - Built with Swift, optimized for Apple Silicon

Requirements

• macOS 13.0 or later
• Apple Silicon (M1/M2/M3) recommended
• Microphone access permission

Installation

Download Pre-built App

Install with Homebrew Cask:

brew install --cask zyuapp/tap/just-speak

Or download the latest zip from the GitHub Releases page and move JustSpeak.app to /Applications.

Note: builds are currently unsigned and not notarized, so macOS may show a security warning on first launch. If that happens, open System Settings > Privacy & Security, scroll to Security, click Open Anyway for JustSpeak, then confirm Open.

Build from Source

# Clone and setup
git clone https://github.com/zyuapp/just-speak.git
cd just-speak
./setup-whisper.sh

# Build
xcodebuild -project JustSpeak.xcodeproj -scheme JustSpeak -configuration Debug clean build

# Find built app
ls -la ~/Library/Developer/Xcode/DerivedData/JustSpeak-*/Build/Products/Debug/JustSpeak.app

Usage

1. First Launch - Grant microphone permission when prompted, download Whisper model from menu
2. Record - Hold Command key (⌘) and speak
3. Optional LLM Setup - Open app menu > Settings... and add Base URL, Model, and API key
4. Transcribe - Release Command to stop and auto-paste
5. Cancel - Press Escape while holding to abort

Tech Stack

• SwiftUI + AppKit - Menu bar app with NSApplicationDelegateAdaptor
• AVFoundation - Native audio recording (16kHz mono WAV)
• Whisper.cpp - Local transcription with Metal acceleration
• Sandboxed - Minimal permissions, secure by design

Development

See CLAUDE.md for architecture details and development guidelines.

Acknowledgments

• Whisper.cpp - High-performance inference of OpenAI's Whisper
• OpenAI Whisper - Robust speech recognition model