VoxTape

Real-time meeting transcription & AI-powered note-taking — 100% local, 100% private

---

A macOS desktop app for real-time meeting transcription (Teams, Meet, Zoom...) with automatic summary and key point generation. Everything runs locally — no external APIs, no data sent to the cloud.

Features

• Real-time transcription — Simultaneous mic + system audio capture (Whisper Turbo)
• Local AI — Summaries, key points, action items via Ministral 3B
• Contextual chat — Ask questions about your past meetings
• Bilingual — French and English (UI, transcription, AI prompts)
• Full-text search — SQLite FTS5
• Export — Markdown or plain text
• 100% offline — No connection required after model download

Requirements

Component	Minimum	Recommended
macOS	14.2 (Sonoma)	15+ (Sequoia)
RAM	16 GB	32 GB
Storage	10 GB	20 GB
Processor	Apple Silicon (M1)	M2/M3/M4

System audio capture requires macOS 14.2+ (ScreenCaptureKit). Intel Macs are not supported.

Installation (Users)

1. Download the DMG from Releases
2. Drag VoxTape into Applications

Gatekeeper bypass

The app is not notarized. macOS will show "app is damaged". Run:

xattr -cr /Applications/VoxTape.app

First launch

1. Open VoxTape
2. The setup wizard downloads AI models (~5 GB)
3. Grant microphone + screen recording access in System Settings

Development

Prerequisites

• Node.js 22+ (recommended: nvm — nvm install 22)
• Rust (required for system audio capture)

  curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
  source "$HOME/.cargo/env"

Quick start

# Clone and install
git clone https://github.com/Lingelo/VoxTape.git
cd VoxTape
npm install

# Build native module (system audio capture)
npm run build:native

# Download AI models (required before first run)
npm run download-model       # STT: Silero VAD + Whisper Turbo (~540 MB)
npm run download-llm-model   # LLM: Ministral 3B Q4_K_M (~2.1 GB)

# Run in dev mode (hot-reload)
npm run dev

The app opens automatically. Angular dev server runs on http://localhost:4200.

Troubleshooting — Electron failed to install correctly or Could not locate the bindings file (better-sqlite3)

Both symptoms mean npm lifecycle scripts were skipped during install, so Electron's binary was never downloaded and better-sqlite3 was never rebuilt against Electron's ABI. The project ships a local .npmrc (ignore-scripts=false) that handles this for most setups. Two cases where it doesn't:

1. You have NPM_CONFIG_IGNORE_SCRIPTS=true in your shell env (common supply-chain hardening). Env vars beat .npmrc per npm config precedence. Reinstall with the CLI flag, which beats env vars:

rm -rf node_modules && npm install --ignore-scripts=false

2. You installed deps before pulling the .npmrc. npm sees "up to date" on the next npm install and doesn't replay the scripts. Force the root postinstall once:

npm run postinstall

This re-runs Electron's binary installer (if missing) and rebuilds better-sqlite3 via electron-rebuild.

macOS Permissions (dev mode)

In dev mode, macOS associates permissions with the Electron binary, not a packaged app. You must manually grant access:

1. Open System Settings > Privacy & Security
2. For both Microphone and Screen Recording:
   • Click +
   • Press Cmd+Shift+G to open the path dialog
   • Navigate to <project-root>/node_modules/electron/dist/Electron.app
   • Add it and ensure the toggle is on
3. Quit and relaunch the app — macOS requires a full process restart after granting Screen Recording

Tip: If permissions aren't working, reset and retry: tccutil reset ScreenCapture then relaunch the app.

Commands

Command	Description
npm run dev	Development mode with hot-reload
npm test	Run tests (Vitest)
npm run build	Production build
npm run package	Create VoxTape.app
npm run make	Create DMG + ZIP
npm run build:native	Compile Rust native module

Architecture

┌─────────────────────────────────────────────────────────────┐
│  Electron Main Process                                      │
│  ┌───────────────────────────────────────────────────────┐  │
│  │  NestJS Backend                                       │  │
│  │  AudioModule → SttModule → stt-worker (sherpa-onnx)   │  │
│  │  LlmModule ────────────→ llm-worker (node-llama-cpp)  │  │
│  │  DatabaseModule (SQLite) | ConfigModule | ExportModule│  │
│  └───────────────────────────────────────────────────────┘  │
│  IPC Hub (contextBridge)                                    │
└─────────────────────────┬───────────────────────────────────┘
                          │
┌─────────────────────────┴───────────────────────────────────┐
│  Renderer Process (Angular 21 SPA)                          │
│  SessionService | AudioCaptureService | LlmService          │
└─────────────────────────────────────────────────────────────┘

Nx monorepo layout

Project	Stack	Description
apps/electron-shell	Electron + Vite	Main process + STT/LLM workers
apps/renderer	Angular 21	User interface
libs/backend	NestJS 11	Services (audio, STT, LLM, DB, export)
libs/shared-types	TypeScript	Interfaces and IPC constants
libs/native-audio-capture	Rust + napi-rs	System audio capture (ScreenCaptureKit)

AI Models

Model	Size	Usage
Silero VAD	2 MB	Voice activity detection
Whisper Turbo (int8)	538 MB	Speech-to-text — large-v3-turbo distilled
Ministral 3B Q4_K_M	2.1 GB	Summaries and chat

Contributing

1. Fork the project
2. Create a branch (git checkout -b feature/my-feature)
3. Commit (git commit -m 'feat: add my feature')
4. Push (git push origin feature/my-feature)
5. Open a Pull Request

License

MIT — See LICENSE