Notes4Me

A privacy-focused macOS menu bar application for recording, transcribing, and summarising meetings using local AI processing.

All data stays on your machine—no cloud uploads, no subscriptions, no privacy compromises.

Overview

Notes4Me is a macOS menu bar application designed for professionals who want to record and transcribe meetings without compromising privacy. Unlike cloud-based solutions, all processing happens locally on your Mac using state-of-the-art open-source AI models.

What makes Notes4Me different:

✅ Privacy-First: Zero data leaves your machine
✅ Cost-Effective: No subscription fees or per-minute charges
✅ Powerful: Uses the same AI models as paid services
✅ Fast: ~5x real-time transcription on Apple Silicon
✅ Intelligent: Generates structured meeting notes with action items

Features

Core Functionality

Audio Recording: Captures system audio from any application (Zoom, Google Meet, Microsoft Teams, etc.)
Speech-to-Text: Converts recordings to accurate text transcripts using whisper.cpp
AI Summarisation: Generates structured meeting notes with:
- Action items and owners
- Key takeaways
- Meeting purpose and context
- Problems, blockers, and solutions discussed
- Next steps

User Experience

Menu Bar Integration: Lightweight tray app that's always accessible
One-Click Recording: Start/stop recording with a single click
Automatic Processing: Optionally transcribe and summarise recordings automatically
Settings UI: Modern, dark-mode-enabled interface for configuration
File Management: Automatic cleanup of old recordings with configurable retention (1-30 days)
Storage Tracking: Monitor disk usage for recordings, transcripts, and notes

Privacy & Performance

100% Local Processing: No cloud APIs, no external services
Fast Transcription: ~5x real-time on Apple Silicon (M1/M2/M3)
Efficient Summarisation: 2-3 minutes for typical meetings
Secure by Design: Electron security best practices with sandboxed renderer

How It Works

Notes4Me uses a three-stage pipeline to transform raw audio into actionable meeting notes:

┌─────────────────┐
│  1. Recording   │  sox captures system audio via BlackHole virtual device
│   (Real-time)   │  Saves to: ~/Documents/MeetingRecordings/recordings/
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│ 2. Transcription│  whisper.cpp converts speech to text locally
│   (~5x speed)   │  Saves to: ~/Documents/MeetingRecordings/processed/
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│ 3. Summarisation│  Ollama (llama3.2) generates structured notes
│   (2-3 minutes) │  Saves to: ~/Documents/MeetingRecordings/processed/
└─────────────────┘

Technology Stack:

Electron 33.x: Cross-platform desktop framework
sox: Reliable audio recording utility
whisper.cpp: Fast, local speech recognition (from OpenAI's Whisper model)
Ollama: Local LLM runtime for AI summarisation
BlackHole: Virtual audio device for system audio capture

System Requirements

Minimum Requirements

Operating System: macOS 11 (Big Sur) or later
Processor: Apple Silicon (M1/M2/M3) or Intel (x86_64)
RAM: 8GB (16GB recommended for faster processing)
Storage: 5GB free space (for dependencies and models)
Node.js: 18.x or later (if building from source)

Recommended Setup

Processor: Apple Silicon (M1 or newer) for optimal performance
RAM: 16GB for smooth multitasking during processing
Storage: 20GB+ for storing recordings before cleanup

Software Dependencies

All dependencies are installed automatically via the setup.sh script:

Homebrew: macOS package manager
sox: Audio recording utility
Ollama: Local LLM runtime
cmake: Build tool for whisper.cpp
whisper.cpp: Compiled from source during setup
BlackHole 2ch: Virtual audio device (manual installation required)

Installation

Quick Start

For experienced users who want to get running immediately:

# 1. Clone the repository
git clone https://github.com/andyj/Notes4Me.git
cd Notes4Me

# 2. Install Node.js dependencies
npm install

# 3. Run automated setup (installs sox, Ollama, whisper.cpp)
./setup.sh

# 4. Install BlackHole (virtual audio device)
# Download from: https://github.com/ExistentialAudio/BlackHole/releases
# Install BlackHole2ch.pkg

# 5. Configure audio routing (see detailed guide below)
# Open Audio MIDI Setup.app → Create Multi-Output Device

# 6. Start Ollama and pull the model
ollama serve &
ollama pull llama3.2

# 7. Run the app
npm start

Detailed Setup Guide

Step 1: Install Homebrew (if not already installed)

Homebrew is a package manager for macOS that we'll use to install dependencies.

# Check if Homebrew is installed
which brew

# If not installed, install it:
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

Verify installation:

brew --version
# Should output: Homebrew 4.x.x

Step 2: Clone the Repository

git clone https://github.com/andyj/Notes4Me.git
cd Notes4Me

Step 3: Install Node.js Dependencies

npm install

This installs Electron and other JavaScript dependencies defined in package.json.

Expected output:

added 267 packages in 15s

Troubleshooting:

If you see permission errors, don't use sudo. Instead, fix npm permissions: npm docs
If npm is not found, install Node.js: brew install node

Step 4: Run the Automated Setup Script

The setup.sh script installs all system dependencies:

chmod +x setup.sh
./setup.sh

What this script does:

Checks for Homebrew
Installs sox (audio recording utility)
Installs Ollama (local LLM runtime)
Installs cmake (build tool)
Clones and compiles whisper.cpp from source
Downloads the whisper base.en model (~75MB)

Expected duration: 5-10 minutes (depending on internet speed and CPU)

Verify installation:

# Check sox
sox --version
# Should output: sox: SoX v14.4.2

# Check Ollama
ollama --version
# Should output: ollama version is 0.x.x

# Check whisper.cpp
ls -la whisper.cpp/build/bin/whisper-cli
# Should show the compiled binary

Troubleshooting:

If the script fails, check the error message and ensure Homebrew is working: brew doctor
If whisper.cpp compilation fails, ensure you have Xcode Command Line Tools: xcode-select --install

Step 5: Install BlackHole (Virtual Audio Device)

BlackHole is a virtual audio driver that allows Notes4Me to capture system audio.

Installation:

Download BlackHole 2ch from: https://github.com/ExistentialAudio/BlackHole/releases
Download the BlackHole2ch.pkg file (NOT the 16ch version)
Open the .pkg file and follow the installer prompts
Restart your Mac after installation (recommended)

Verify installation:

system_profiler SPAudioDataType | grep BlackHole
# Should output: BlackHole 2ch

Step 6: Configure Audio Routing

This is the most critical step for recording system audio. You need to create a "Multi-Output Device" that sends audio to both your speakers AND BlackHole.

Step-by-step:

Open Audio MIDI Setup.app
- Press Cmd + Space and type "Audio MIDI Setup"
- Or navigate to /Applications/Utilities/Audio MIDI Setup.app
Create a Multi-Output Device:
- Click the + button in the bottom-left corner
- Select "Create Multi-Output Device"
Configure the Multi-Output Device:
- In the right panel, check the boxes for:
  - ✅ Your speakers/headphones (e.g., "MacBook Pro Speakers" or "External Headphones")
  - ✅ BlackHole 2ch
- Important: Your speakers should be FIRST in the list (drag to reorder if needed)
Set as System Default:
- Right-click the Multi-Output Device
- Select "Use This Device For Sound Output"
- Or go to System Settings → Sound → Output and select the Multi-Output Device
Test Audio Routing:
- Play a YouTube video or music
- You should hear audio normally through your speakers
- Notes4Me will now be able to record this audio

Why is this needed?

macOS doesn't allow apps to directly capture system audio for privacy reasons
BlackHole creates a virtual audio device that acts as a "loopback"
The Multi-Output Device sends audio to BOTH your speakers (so you hear it) AND BlackHole (so Notes4Me can record it)

Troubleshooting:

No sound after setup: Ensure your speakers are checked and FIRST in the Multi-Output Device list
BlackHole not visible: Restart your Mac after installing BlackHole
Can't set as default: Use System Settings → Sound → Output instead of Audio MIDI Setup

Step 7: Start Ollama and Download the AI Model

Ollama is the local LLM runtime that powers meeting summarisation.

Start Ollama server:

ollama serve

This starts the Ollama API server at http://localhost:11434. Keep this terminal window open.

In a NEW terminal window, download the AI model:

ollama pull llama3.2

This downloads the llama3.2 model (~2GB). This is a one-time download.

Expected output:

pulling manifest
pulling 8eeb52dfb3bb... 100% ▕████████████▏ 2.0 GB
pulling 73b313b5552d... 100% ▕████████████▏  11 KB
pulling 0ba8f0e314b4... 100% ▕████████████▏  12 KB
pulling 56bb8bd477a5... 100% ▕████████████▏  485 B
success

Verify Ollama is running:

curl http://localhost:11434/api/tags

Should return a JSON response with available models.

Troubleshooting:

If ollama serve fails, try: brew services start ollama (runs as background service)
If port 11434 is in use, stop other services: lsof -ti:11434 | xargs kill

Step 8: Run Notes4Me

You're ready to launch the app!

npm start

You should see:

A menu bar icon appear in the top-right of your screen
The Electron app window may briefly appear and hide
Click the menu bar icon to access recording controls

First-time launch checklist:

Menu bar icon appears: ✅
Click icon → "Start Recording" option visible: ✅
Click "Settings..." → System Status shows all dependencies green: ✅

Audio Routing Configuration

Understanding the Audio Flow

┌─────────────────┐
│  Your Meeting   │  (Zoom, Google Meet, etc.)
│  Application    │
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│ macOS Audio     │  System audio output
│    System       │
└────────┬────────┘
         │
         ├─────────────────────────────┐
         │                             │
         ▼                             ▼
┌─────────────────┐          ┌─────────────────┐
│  Your Speakers  │          │  BlackHole 2ch  │
│  (you hear it)  │          │ (app records it)│
└─────────────────┘          └────────┬────────┘
                                      │
                                      ▼
                              ┌─────────────────┐
                              │   Notes4Me      │
                              │  (via sox)      │
                              └─────────────────┘

Common Audio Routing Issues

Problem: No audio in recording (silent WAV file)

Solution:

Verify Multi-Output Device is selected as system output:

# Check current output device
system_profiler SPAudioDataType | grep "Default Output"

Play test audio (YouTube video) and verify you can hear it
Check that both speakers and BlackHole are checked in Multi-Output Device
Restart the app and try recording again

Problem: Can't hear audio after setting up Multi-Output Device

Solution:

Open Audio MIDI Setup
Select the Multi-Output Device
Ensure your speakers are:
- ✅ Checked (enabled)
- 📍 First in the list (drag to reorder)
- 🔊 Not muted (check system volume)

Problem: Recording has echo or distortion

Solution:

Check sample rate consistency:
- Open Audio MIDI Setup
- Select BlackHole 2ch → Format → 48000.0 Hz
- Select Multi-Output Device → Format → 48000.0 Hz
- Select your speakers → Format → 48000.0 Hz
Restart the recording

Usage

Starting the App

npm start

The app runs in the menu bar (top-right corner of your screen). Click the icon to see:

Start Recording - Begin capturing audio
Settings... - Open settings window
Quit - Exit the app

Recording Your First Meeting

Before the meeting:
- Ensure Ollama is running: ollama serve
- Verify Multi-Output Device is your system output
- Start Notes4Me: npm start
During the meeting:
- Click menu bar icon → "Start Recording"
- Join your meeting (Zoom, Google Meet, etc.)
- The menu bar icon will update to show recording status
- Speak and interact normally
After the meeting:
- Click menu bar icon → "Stop Recording"
- The app will automatically:
  - Stop the audio capture
  - Transcribe the recording (~5x real-time)
  - Generate meeting notes (2-3 minutes)
- Menu bar icon will show progress
View your results:
- Open ~/Documents/MeetingRecordings/processed/
- Find files with today's date (e.g., 2025-11-19_14-30-00_transcript.txt)

Understanding the Output

Notes4Me generates three types of files:

1. Recording (WAV Audio)

Location: ~/Documents/MeetingRecordings/recordings/
Format: YYYY-MM-DD_HH-MM-SS_recording.wav
Size: ~10MB per minute (48kHz, 16-bit stereo)
Retention: Auto-deleted after 7 days (configurable)

Example:

2025-11-19_14-30-00_recording.wav  (300 MB for 30-min meeting)

2. Transcript (Plain Text)

Location: ~/Documents/MeetingRecordings/processed/
Format: YYYY-MM-DD_HH-MM-SS_transcript.txt
Content: Raw speech-to-text output from whisper.cpp
Retention: Kept indefinitely

Example:

Welcome everyone to today's standup meeting. Let's go around and share what
we've been working on. Sarah, would you like to start?

Sure. Yesterday I finished the authentication refactor and pushed it to the
dev branch. Today I'm going to work on the user settings page...

3. Meeting Notes (Markdown)

Location: ~/Documents/MeetingRecordings/processed/
Format: YYYY-MM-DD_HH-MM-SS_notes.md
Content: AI-generated structured summary
Retention: Kept indefinitely

Example structure:

# Meeting Notes

## Action Items
- [ ] Sarah: Complete user settings page by Friday
- [ ] John: Review authentication PR #234
- [ ] Team: Test new login flow in staging

## Meeting Purpose
Daily standup to sync on sprint progress and blockers

## Key Takeaways
- Authentication refactor is complete and ready for review
- User settings page on track for Friday delivery
- Staging environment needs to be updated before testing

## Topics Discussed

### Problems
- Staging environment is outdated (2 weeks behind main)

### Blockers
- Waiting for design team to finalize settings page mockups

### Solutions
- DevOps to update staging tonight
- Using existing design patterns for settings page

## Next Steps
- Deploy staging environment update
- Review and merge authentication PR
- Begin testing new login flow

Settings & Configuration

Open settings by clicking the menu bar icon → "Settings..."

System Status Tab

View real-time dependency status (sox, whisper, Ollama, BlackHole)
Click "Refresh Status" to re-check dependencies
Green checkmarks = ready, Red X = needs attention

Settings Tab

Output Directory: Change where recordings are saved (default: ~/Documents/MeetingRecordings)
Retention Period: How many days to keep recordings (1-30 days, default: 7)
Auto-Process: Automatically transcribe and summarise after recording stops

Recordings Tab

View all recordings with metadata (date, size, processing status)
Actions per recording:
- 📂 Open: View in Finder
- ⚙️ Process: Manually transcribe/summarise
- 🗑️ Delete: Remove recording and associated files

Storage Tab

View total storage usage
See counts: recordings, transcripts, notes
Cleanup Old Recordings: Manually trigger retention policy
Open Output Folder: Quick access to recordings directory

Performance

Benchmarks (M1 MacBook Air, 8GB RAM)

Recording Length	Transcription Time	Summarisation Time	Total Processing Time
15 minutes	~3 minutes	~1.5 minutes	~4.5 minutes
30 minutes	~6 minutes	~2 minutes	~8 minutes
60 minutes	~12 minutes	~3 minutes	~15 minutes

Performance Notes

Transcription: whisper.cpp base.en model processes at ~5x real-time on Apple Silicon
Summarisation: llama3.2 generates ~50 tokens/second
Processing: Runs in background, doesn't block other work
Memory: ~2GB RAM during transcription, ~4GB during summarisation

Optimization Tips

Use base.en model: Faster than multilingual models, sufficient for English
Close other apps: Gives more CPU/RAM to processing
Disable auto-process: Process recordings manually during idle time
SSD recommended: Faster file I/O for large WAV files

Troubleshooting

Installation Issues

"Homebrew not found"

# Install Homebrew
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# Add to PATH (Apple Silicon)
echo 'eval "$(/opt/homebrew/bin/brew shellenv)"' >> ~/.zprofile
eval "$(/opt/homebrew/bin/brew shellenv)"

"npm: command not found"

# Install Node.js via Homebrew
brew install node

# Verify installation
node --version
npm --version

"whisper.cpp compilation failed"

# Install Xcode Command Line Tools
xcode-select --install

# Install cmake
brew install cmake

# Retry setup
./setup.sh

Audio Recording Issues

"BlackHole not found in audio devices"

# Verify BlackHole is installed
system_profiler SPAudioDataType | grep -i blackhole

# If not found, reinstall BlackHole:
# 1. Download from https://github.com/ExistentialAudio/BlackHole/releases
# 2. Install BlackHole2ch.pkg
# 3. Restart your Mac

"Recording is silent (0 bytes or no speech detected)"

Common causes:

Multi-Output Device not set as system output
BlackHole not included in Multi-Output Device
No audio playing during recording

Solution:

# 1. Check current output device
system_profiler SPAudioDataType | grep "Default Output"

# 2. Re-configure Multi-Output Device:
#    - Open Audio MIDI Setup
#    - Verify both speakers and BlackHole 2ch are checked
#    - Set as system default output

# 3. Test with audio playback:
#    - Play a YouTube video
#    - Start recording
#    - Stop after 10 seconds
#    - Check file size: ls -lh ~/Documents/MeetingRecordings/recordings/
#    - Should be ~1MB for 10 seconds

"Can't hear audio after setup"

# Verify speakers are enabled in Multi-Output Device
# 1. Open Audio MIDI Setup
# 2. Select Multi-Output Device
# 3. Ensure your speakers are:
#    - Checked (✓)
#    - First in the list (drag to reorder)
# 4. Check system volume (not muted)

Transcription Issues

"whisper.cpp binary not found"

# Verify binary exists
ls -la whisper.cpp/build/bin/whisper-cli

# If not found, recompile
cd whisper.cpp
rm -rf build
cmake -B build
cmake --build build
cd ..

"Whisper model not found"

# Download base.en model manually
cd whisper.cpp
bash ./models/download-ggml-model.sh base.en
ls -lh models/ggml-base.en.bin  # Should be ~75MB
cd ..

"Transcription takes too long"

Normal: 30-min recording → 6 min transcription If slower:

Close other apps to free up CPU
Check Activity Monitor for high CPU usage
Consider using a smaller model (tiny.en) for faster processing:
```
cd whisper.cpp
bash ./models/download-ggml-model.sh tiny.en
```
Then update services/transcriber.js to use ggml-tiny.en.bin

Summarisation Issues

"Ollama not available"

# Check if Ollama is running
curl http://localhost:11434/api/tags

# If connection refused, start Ollama
ollama serve

# Or run as background service
brew services start ollama

"llama3.2 model not found"

# List installed models
ollama list

# If llama3.2 not listed, download it
ollama pull llama3.2

# Verify download
ollama list | grep llama3.2

"Summarisation fails or produces gibberish"

Possible causes:

Empty or very short transcript
Ollama out of memory
Model corruption

Solution:

# 1. Check transcript has content
cat ~/Documents/MeetingRecordings/processed/*_transcript.txt | wc -w
# Should have 100+ words for meaningful summary

# 2. Restart Ollama
pkill ollama
ollama serve

# 3. Re-download model if corrupted
ollama rm llama3.2
ollama pull llama3.2

General Issues

"App crashes on startup"

# Check logs
npm start 2>&1 | tee app.log

# Look for specific error messages in app.log

"High CPU usage"

During recording: Normal (sox process) During transcription: Normal (whisper process using 100% CPU) Idle: Should be <5% CPU

If high CPU when idle:

# Check for stuck processes
ps aux | grep -E "(sox|whisper|ollama)"

# Kill stuck processes
pkill sox
pkill whisper-cli

"Storage filling up"

# Check storage usage
du -sh ~/Documents/MeetingRecordings/recordings/
du -sh ~/Documents/MeetingRecordings/processed/

# Manual cleanup (removes recordings older than 7 days)
find ~/Documents/MeetingRecordings/recordings/ -name "*.wav" -mtime +7 -delete

# Or use Settings UI:
# Click "Cleanup Old Recordings" button

Privacy & Security

Data Collection

Notes4Me collects ZERO data. Everything happens locally on your Mac.

❌ No analytics or telemetry
❌ No cloud uploads
❌ No external API calls (except to localhost:11434 for Ollama)
❌ No user accounts or authentication
❌ No network requests to third-party servers

Data Storage

All data is stored in: ~/Documents/MeetingRecordings/

Permissions:

Files are owned by your user account
Standard macOS file permissions (read/write for owner only)
No root or elevated privileges required

Retention Policy:

Recordings (WAV): Auto-deleted after 7 days (configurable: 1-30 days)
Transcripts (TXT): Kept indefinitely (manual deletion required)
Notes (MD): Kept indefinitely (manual deletion required)

Network Security

Notes4Me only makes network requests to:

http://localhost:11434 - Ollama API (local server, not internet)

You can verify this by running the app with network disconnected - it will work normally (assuming Ollama is already running).

Code Security

Sandboxed renderer: Electron security best practices
No nodeIntegration: Renderer process can't access Node.js APIs
contextBridge: Secure IPC communication only
No eval(): No dynamic code execution
Open source: All code is auditable in this repository

Threat Model

Protects against:

✅ Third-party data breaches (no cloud storage)
✅ Network interception (no external requests)
✅ Vendor lock-in (open source, local data)
✅ Subscription price increases (free forever)

Does NOT protect against:

❌ Physical access to your Mac (files are not encrypted at rest)
❌ Malware on your system (standard macOS security applies)
❌ Backup leaks (if you back up ~/Documents to cloud services)

Recommendations for maximum privacy:

Use FileVault (full-disk encryption)
Exclude ~/Documents/MeetingRecordings/ from cloud backups (iCloud, Dropbox, etc.)
Manually delete sensitive recordings after reviewing notes

FAQ

General

Q: Does this work on Windows or Linux? A: Currently macOS only. Windows/Linux support is on the roadmap but requires significant changes (different audio capture methods, virtual audio drivers, etc.).

Q: Does it work on Intel Macs? A: Yes, but performance is slower (~3x real-time transcription vs 5x on Apple Silicon).

Q: Can I use this commercially? A: Yes! MIT license allows commercial use. However, check the licenses of dependencies (whisper.cpp, Ollama) if redistributing.

Q: How much does it cost? A: Free forever. No subscriptions, no API fees, no hidden costs. You only need a Mac and disk space.

Recording

Q: Can I record Zoom/Google Meet/Microsoft Teams? A: Yes, as long as you've set up the Multi-Output Device correctly. The app records all system audio.

Q: Can I record my microphone input (what I'm saying)? A: Not currently. Notes4Me records system OUTPUT (what you hear), not microphone INPUT. This is a limitation we plan to address.

Q: What's the maximum recording length? A: No hard limit, but consider:

Disk space: ~10MB per minute (~600MB for 1 hour)
Processing time: ~15 minutes for 1-hour recording
Memory: Large files may cause issues on 8GB RAM systems

Q: Can I pause and resume a recording? A: Not currently. Start a new recording for each distinct session.

Transcription

Q: How accurate is the transcription? A: Very accurate for clear English speech. Whisper base.en model has ~5% word error rate on clean audio. Accuracy degrades with:

Heavy accents
Background noise
Multiple speakers talking over each other
Poor audio quality (low-bitrate conferencing apps)

Q: Can I transcribe other languages? A: Not with the default base.en model. You can download multilingual models:

cd whisper.cpp
bash ./models/download-ggml-model.sh base  # Multilingual

Then update services/transcriber.js to use ggml-base.bin

Q: Can I transcribe existing audio files? A: Not directly in the UI, but you can manually:

cd whisper.cpp
./build/bin/whisper-cli -m models/ggml-base.en.bin -f /path/to/your/audio.wav -otxt

Summarisation

Q: Can I customise the summary format? A: Yes! Edit the prompt in services/summariser.js (line ~60) to change the structure, add custom sections, or adjust tone.

Q: Can I use a different AI model? A: Yes! Notes4Me works with any Ollama-compatible model:

ollama pull mistral        # Faster, less detailed
ollama pull llama3.1:70b   # Slower, more detailed (requires 64GB+ RAM)

Then update services/summariser.js to use your preferred model.

Q: Do summaries include speaker identification? A: Not currently. Whisper doesn't do speaker diarisation. This is a planned feature.

Privacy

Q: Can someone intercept my recordings over the network? A: No. All processing happens locally. The only network request is to localhost:11434 (Ollama), which never leaves your machine.

Q: Are recordings encrypted? A: No, files are stored as plain WAV/TXT/MD. For encryption at rest, use FileVault (macOS full-disk encryption).

Q: Can I delete recordings automatically after processing? A: Yes! Enable "Auto-Process" in settings, then set retention period to 1 day. Recordings will be deleted 24 hours after creation.

Performance

Q: Why is transcription slower than advertised? A: Performance varies by:

CPU model (M1 vs M2 vs Intel)
Available RAM (8GB vs 16GB+)
System load (other apps using CPU)
Audio quality (noisy audio takes longer)

Q: Can I speed up processing? A: Yes:

Use a smaller whisper model (tiny.en instead of base.en)
Close other apps to free up CPU
Disable auto-process and run manually during idle time

Q: Does it use GPU acceleration? A: whisper.cpp uses Apple's Metal framework on Apple Silicon for some operations, but primarily uses CPU.

Development

Project Structure

Notes4Me/
├── main.js                 # Electron main process
│                           #  - Creates tray icon
│                           #  - Orchestrates recording pipeline
│                           #  - Manages IPC communication
│
├── preload.js              # Secure IPC bridge
│                           #  - Exposes safe API to renderer
│                           #  - Uses contextBridge
│
├── package.json            # Node.js dependencies & scripts
├── .gitignore              # Git exclusions
├── LICENSE                 # MIT License
├── README.md               # This file
├── setup.sh                # Automated dependency installer
│
├── services/               # Core business logic
│   ├── recorder.js         #  - Audio capture via sox
│   ├── transcriber.js      #  - whisper.cpp integration
│   ├── summariser.js       #  - Ollama API client
│   └── fileManager.js      #  - File operations & cleanup
│
├── utils/                  # Utilities
│   └── audioDevices.js     #  - BlackHole detection
│
├── renderer/               # Settings window UI
│   ├── index.html          #  - Settings window structure
│   ├── styles.css          #  - Dark mode styling
│   └── app.js              #  - Frontend logic & IPC calls
│
├── assets/                 # Application assets
│   ├── icon.png            #  - Menu bar icon (TODO)
│   └── README.md           #  - Asset documentation
│
├── models/                 # Whisper models
│   └── README.md           #  - Model documentation
│
└── whisper.cpp/            # Speech recognition (git submodule)
    └── ...                 #  - Compiled during setup

Development Commands

# Install dependencies
npm install

# Run in development mode (with DevTools)
npm start

# Build for distribution
npm run build

# Rebuild native modules for Electron
npm run postinstall

# Clean build artifacts
rm -rf dist/ build/ out/

Architecture Overview

Main Process (main.js)

Tray Icon: Creates menu bar icon and context menu
IPC Handlers: Responds to renderer requests (startRecording, stopRecording, etc.)
Pipeline Orchestration: Chains recorder → transcriber → summariser
Window Management: Creates and manages settings window

Services Layer

Recorder: Spawns sox subprocess, monitors file size, handles cleanup
Transcriber: Spawns whisper.cpp, parses progress, returns transcript path
Summariser: Streams from Ollama API, formats markdown output
FileManager: CRUD operations, retention policy, storage stats

Renderer Process (renderer/)

Settings UI: Modern dark-mode interface
IPC Client: Calls main process via window.meetingRecorder API
State Management: Local component state (no framework)

Building from Source

# Clone with submodules
git clone --recurse-submodules https://github.com/andyj/Notes4Me.git
cd Notes4Me

# Install and setup
npm install
./setup.sh

# Run
npm start

Creating a Distribution Build

Local Build

npm run build

This uses electron-builder to create a macOS .app bundle in dist/.

Output:

dist/Notes4Me-1.0.0-arm64.dmg - DMG installer (~222MB)
dist/Notes4Me-1.0.0-arm64-mac.zip - ZIP archive (~218MB)

Customization:

Edit package.json → build section for app metadata
Add app icon: assets/icon.png (1024x1024px)
Code signing: Set APPLE_ID and APPLE_ID_PASSWORD environment variables

Automated Builds (GitHub Actions)

Every push to main automatically builds the app using GitHub Actions:

What happens:

Code is pushed to GitHub
GitHub Actions runner (macOS) checks out code
Installs Node.js dependencies
Runs npm run build
Uploads DMG and ZIP as downloadable artifacts

Download builds:

Go to: https://github.com/andyj/Notes4Me/actions
Click on latest workflow run
Download artifacts from the "Artifacts" section
Artifacts are kept for 30 days

Creating releases: To create a GitHub Release with downloadable installers:

# Tag a release
git tag v1.0.0
git push origin v1.0.0

This will:

Build the app automatically
Create a GitHub Release at https://github.com/andyj/Notes4Me/releases
Attach DMG and ZIP files to the release
Generate release notes from commits

Workflow file: .github/workflows/build.yml

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for guidelines.

Areas for contribution:

Windows/Linux support
Speaker diarisation (who said what)
Real-time transcription
Custom AI prompts
Export formats (PDF, DOCX)
Improved error handling
Unit tests

Development setup:

Fork the repository
Create a feature branch: git checkout -b feature/my-feature
Make changes and test thoroughly
Commit with clear messages: git commit -m "Add speaker diarisation"
Push and create a Pull Request

License

MIT License - see LICENSE file for details.

Summary:

✅ Commercial use
✅ Modification
✅ Distribution
✅ Private use
❌ Liability
❌ Warranty

Acknowledgments

This project stands on the shoulders of giants:

whisper.cpp - Georgi Gerganov's incredible C++ implementation of OpenAI's Whisper
Ollama - Making local LLMs accessible to everyone
BlackHole - Virtual audio driver that makes macOS audio routing possible
sox - The Swiss Army knife of audio manipulation
Electron - Cross-platform desktop framework

Support

Issues & Bug Reports:

GitHub Issues: https://github.com/andyj/Notes4Me/issues

Discussions & Questions:

GitHub Discussions: https://github.com/andyj/Notes4Me/discussions

Security Vulnerabilities:

Email: security@andyjarrett.com (or open a private security advisory)

Changelog

v1.0.0 (Initial Release)

Core recording, transcription, and summarisation pipeline
macOS menu bar integration
Settings UI with dark mode
Automatic file cleanup
Privacy-first local processing

Built with ❤️ for privacy-conscious professionals

Name		Name	Last commit message	Last commit date
Latest commit History 18 Commits
.github/workflows		.github/workflows
assets		assets
models		models
renderer		renderer
services		services
utils		utils
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
main.js		main.js
package-lock.json		package-lock.json
package.json		package.json
preload.js		preload.js
setup.sh		setup.sh
summarise-only.js		summarise-only.js

License

andyj/Notes4Me

Folders and files

Latest commit

History

Repository files navigation

Notes4Me

Table of Contents

Overview

Features

Core Functionality

User Experience

Privacy & Performance

How It Works

System Requirements

Minimum Requirements

Recommended Setup

Software Dependencies

Installation

Quick Start

Detailed Setup Guide

Step 1: Install Homebrew (if not already installed)

Step 2: Clone the Repository

Step 3: Install Node.js Dependencies

Step 4: Run the Automated Setup Script

Step 5: Install BlackHole (Virtual Audio Device)

Step 6: Configure Audio Routing

Step 7: Start Ollama and Download the AI Model

Step 8: Run Notes4Me

Audio Routing Configuration

Understanding the Audio Flow

Common Audio Routing Issues

Usage

Starting the App

Recording Your First Meeting

Understanding the Output

1. Recording (WAV Audio)

2. Transcript (Plain Text)

3. Meeting Notes (Markdown)

Settings & Configuration

System Status Tab

Settings Tab

Recordings Tab

Storage Tab

Performance

Benchmarks (M1 MacBook Air, 8GB RAM)

Performance Notes

Optimization Tips

Troubleshooting

Installation Issues

"Homebrew not found"

"npm: command not found"

"whisper.cpp compilation failed"

Audio Recording Issues

"BlackHole not found in audio devices"

"Recording is silent (0 bytes or no speech detected)"

"Can't hear audio after setup"

Transcription Issues

"whisper.cpp binary not found"

"Whisper model not found"

"Transcription takes too long"

Summarisation Issues

"Ollama not available"

"llama3.2 model not found"

"Summarisation fails or produces gibberish"

General Issues

"App crashes on startup"

"High CPU usage"

"Storage filling up"

Privacy & Security

Data Collection

Data Storage

Network Security

Code Security

Threat Model

FAQ

General

Recording

Transcription

Summarisation

Packages