PrivaNote 🔒

Privacy-First AI Meeting Assistant - Local transcription and intelligent analysis with dual-mode AI processing.

🌟 Features

• 🎙️ Local Audio Transcription: Powered by faster-whisper, processes audio entirely on your device
• 🤖 Dual-Mode AI Analysis: Choose between OpenAI (cloud) or local Gemma 3n models via Ollama
• 🔒 Complete Privacy Control: Audio files never leave your device, choose your privacy level
• 📊 Smart Meeting Intelligence: Automatic summaries, action items, key decisions, and topic extraction
• 📁 Meeting Archive: Searchable history with export to Markdown, JSON, and other formats
• 🌐 Web-Based Interface: Modern, responsive UI accessible from any browser
• ⚡ Cross-Platform: Runs on Windows, macOS, and Linux with automatic device optimization

🛡️ Privacy Modes

Mode	Description	Privacy Level	Audio Processing	AI Analysis
🏠 Local Gemma	Complete privacy	🟢 Maximum	Local (Whisper)	Local (Ollama/Gemma)
☁️ OpenAI Cloud	High-quality analysis	🟡 Moderate	Local (Whisper)	Cloud (OpenAI API)
🔧 Basic Analysis	Keyword-based	🟢 Maximum	Local (Whisper)	Local (Keywords)

🔒 Your audio files never leave your device regardless of the AI mode you choose.

🚀 Quick Start

Option 1: Run on Replit (Easiest)

1. Fork this Replit or run directly in the browser
2. Add OpenAI API Key (optional): Go to Secrets tab → Add OPENAI_API_KEY
3. Click Run - The app will start automatically on port 5000
4. Upload audio and start analyzing your meetings!

Option 2: Local Installation (Maximum Privacy)

Prerequisites

• Python 3.11 or 3.12 (Python 3.13 is not supported due to the removal of the audioop module)
• Git
• FFmpeg (required for audio processing)

Installation Steps

1. Install FFmpeg (required for audio processing)

   Linux (Ubuntu/Debian):

   sudo apt update
   sudo apt install ffmpeg

   Linux (CentOS/RHEL/Fedora):

   # For CentOS/RHEL
   sudo yum install ffmpeg
   
   # For Fedora
   sudo dnf install ffmpeg

   macOS:

   # Using Homebrew
   brew install ffmpeg
   
   # Using MacPorts
   sudo port install ffmpeg

   Windows:

   # Using Chocolatey
   choco install ffmpeg
   
   # Using Winget
   winget install ffmpeg

   Or download manually from https://ffmpeg.org/download.html and add to PATH.

   Verify installation:

   ffmpeg -version
   ffprobe -version

2. Clone the repository

   git clone https://github.com/Rebell-Leader/PrivaNote.git
   cd PrivaNote

3. Install dependencies

   Using uv (recommended):

   pip install uv
   uv sync

   Using pip:

   pip install streamlit>=1.48.0 faster-whisper>=1.2.0 openai>=1.99.1 ollama>=0.5.2 pandas>=2.3.1 pydub>=0.25.1 torch>=2.8.0 torchaudio>=2.8.0

4. Run the application

   streamlit run app.py --server.port 5000

5. Open your browser to http://localhost:5000

🏠 For Maximum Privacy: Local Ollama Setup

To enable completely local AI processing with Gemma 3n models:

1. Install Ollama

   # Windows/Mac: Download from https://ollama.com
   # Linux:
   curl -fsSL https://ollama.com/install.sh | sh

2. Install Gemma 3 models

   # Recommended 4B model (balanced performance/quality)
   ollama pull gemma3
   
   # Alternative options:
   ollama pull gemma3:1b     # Faster, smaller (1.7GB)
   ollama pull gemma3:27b    # Highest quality (requires 16GB+ RAM)

3. Restart the app - Ollama integration will be automatically detected

📋 Usage Guide

Basic Workflow

1. Upload Audio: Support for WAV, MP3, MP4, M4A, FLAC, OGG formats
2. Choose AI Provider: Select your preferred privacy/quality balance
3. Review Results: Automatically generated summaries, action items, and insights
4. Export & Archive: Save to Markdown, search previous meetings

AI Provider Selection

• OpenAI (Cloud): Requires API key, highest quality analysis
• Local Gemma: Requires Ollama setup, complete privacy
• Basic Analysis: No setup required, keyword-based extraction

Supported Audio Formats

• WAV, MP3, MP4, M4A, FLAC, OGG
• Automatic format conversion for optimal Whisper compatibility
• Works with virtual meeting recordings (Zoom, Teams, etc.)

🏗️ Technical Architecture

Core Components

• Frontend: Streamlit web application with responsive design
• Audio Processing: PyDub for format conversion and metadata extraction
• Transcription: faster-whisper for local speech-to-text
• AI Analysis: Dual-mode system (OpenAI API + Ollama integration)
• Storage: Local browser session with export capabilities

Technology Stack

• Python 3.11+ - Core runtime
• Streamlit - Web framework and UI
• faster-whisper - Local speech recognition
• OpenAI API - Cloud AI analysis (optional)
• Ollama + Gemma 3n - Local AI analysis (optional)
• PyTorch - ML framework with CPU/GPU optimization

Privacy by Design

• Local-first processing: Audio transcription happens on your device.
• Configurable AI providers: You choose your desired level of privacy.
• Ephemeral by Default: Meeting data is held only in your browser's session memory and is discarded when you close the tab.
• User-Controlled Persistence: You have full control over what data to save, via manual exports to Markdown or other formats.

🎯 Use Cases

👔 For Business Professionals

• Meetings: Capture decisions, action items, and follow-ups
• Interviews: Generate structured candidate assessments
• Client Calls: Track requirements and commitments
• Team Standups: Document progress and blockers

🚀 For Founders & Entrepreneurs

• Investor Meetings: Capture feedback and next steps
• Team Updates: Track project progress and decisions
• Customer Discovery: Organize insights and pain points
• Board Meetings: Generate structured reports and summaries

🏢 For Enterprise Teams

• Compliance: Maintain meeting records with privacy controls
• Knowledge Management: Build searchable meeting archives
• Remote Work: Enhance async communication with meeting summaries
• Training: Create documentation from recorded sessions

🔧 Configuration

Environment Variables

# Optional: OpenAI API key for cloud analysis
OPENAI_API_KEY=your_openai_api_key

# Optional: Custom Ollama host (default: localhost:11434)
OLLAMA_HOST=localhost:11434

Streamlit Configuration

The app includes optimized settings in .streamlit/config.toml:

• Headless mode for deployment
• Proper host binding (0.0.0.0:5000)
• Performance optimizations

🐛 Troubleshooting

Common Issues

Whisper initialization fails:

• The app automatically tries multiple device configurations
• CPU mode with int8 compute type should work on any system
• Check error messages in the console for specific issues

OpenAI API errors:

• Verify your API key is correctly set in environment variables
• Ensure you have sufficient API credits
• Check your internet connection

Ollama not detected:

• Make sure Ollama service is running (ollama serve)
• Verify models are installed (ollama list)
• Check if port 11434 is accessible

Audio upload issues:

• Supported formats: WAV, MP3, MP4, M4A, FLAC, OGG
• Maximum file size depends on available memory
• Try converting audio to WAV format if issues persist

Performance Tips

• CPU-only systems: Use Whisper's int8 mode (automatic)
• GPU systems: CUDA acceleration (automatic detection)
• Large files: Consider splitting into smaller segments
• Memory usage: Close unused browser tabs during processing

🤝 Contributing

We welcome contributions! Areas where you can help:

• 🐛 Bug fixes - Report issues and submit fixes
• 🎨 UI/UX improvements - Enhance the user interface
• 🔌 Integrations - Add support for more AI models or export formats
• 📚 Documentation - Improve setup guides and usage examples
• 🧪 Testing - Test across different platforms and configurations

Development Setup

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Test thoroughly
5. Submit a pull request

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

🗺️ Roadmap

🎯 Near Term

• Speaker diarization - Identify individual speakers
• Real-time processing - Live meeting transcription
• Mobile support - Responsive design improvements
• API integrations - Notion, Confluence, Slack export

🚀 Future Vision

• Native mobile apps - iOS and Android with local AI
• Team collaboration - Shared meeting spaces
• Advanced search - Semantic search across meeting history
• Custom models - Fine-tuned models for specific domains
• Workflow automation - Trigger actions based on meeting content

💬 Support & Community

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Documentation: This README and inline help
• Email: Contact us

🙏 Acknowledgments

• OpenAI - Whisper model and API services
• Google - Gemma 3n models and research
• Ollama - Local model serving platform
• Streamlit - Web application framework
• faster-whisper - Optimized Whisper implementation

---

Built with ❤️ for privacy-conscious professionals who want AI-powered meeting intelligence without compromising their data security.