Skip to content

replit.md

Latest commit

 

History

History
126 lines (103 loc) · 6.84 KB

replit.md

File metadata and controls

126 lines (103 loc) · 6.84 KB

PrivaNote - Privacy-First AI Meeting Assistant

Overview

PrivaNote is a privacy-focused meeting assistant application built with Streamlit that provides local audio transcription and intelligent analysis capabilities. The application processes meeting recordings entirely on the user's device, ensuring complete data privacy while offering comprehensive meeting analysis features including summaries, action item extraction, and searchable meeting archives.

The system is designed around a modular architecture with separate services for audio processing, transcription, AI analysis, storage, and export functionality. It leverages local AI models (Whisper for transcription and planned integration with Gemma 3n E4B for analysis) to maintain privacy while providing enterprise-grade meeting intelligence.

User Preferences

Preferred communication style: Simple, everyday language. Vision: Cross-platform, local AI meeting assistant using Gemma 3n E4B and Whisper for privacy-first intelligent transcription and analysis.

System Architecture

Frontend Architecture

  • Framework: Streamlit web application with responsive layout
  • State Management: Session-based storage using Streamlit's built-in session state
  • UI Components: Wide layout with expandable sidebar and privacy-focused design
  • Caching: Resource caching for service initialization to improve performance

Backend Architecture

  • Modular Service Design: Five core services handle distinct responsibilities:
    • AudioProcessor: Handles file format conversion and metadata extraction
    • TranscriptionService: Local audio-to-text conversion using faster-whisper
    • AIAnalysisService: Meeting content analysis (currently OpenAI API, planned Gemma integration)
    • StorageService: Local data persistence in browser session
    • ExportService: Multi-format export capabilities (Markdown, JSON planned)

Audio Processing Pipeline

  • Input Support: Multiple audio formats (WAV, MP3, MP4, M4A, FLAC, OGG)
  • Preprocessing: Automatic conversion to WAV format for optimal Whisper compatibility
  • Metadata Extraction: Duration, file size, sample rate, and channel information
  • Quality Optimization: Format standardization for consistent transcription results

Transcription Architecture

  • Local Processing: faster-whisper implementation for on-device transcription
  • Model Management: Configurable model sizes with automatic device detection
  • Performance Optimization: CPU/GPU automatic selection with appropriate compute types
  • Language Support: Automatic language detection with manual override capability

AI Analysis System

  • Triple-Mode Architecture: Three AI processing options for different privacy and performance needs
    • OpenAI (Cloud): High-quality cloud analysis via gpt-4o API
    • Ollama (Local): Privacy-first local processing with Gemma 3n models
    • LM Studio (Local Server): OpenAI-compatible local server with custom configuration
  • API Configuration: Uses OPENAI_API_KEY environment variable for cloud mode
  • LM Studio Integration: Configurable host/port/model with real-time connection testing
  • Analysis Capabilities:
    • Meeting summarization
    • Action item extraction
    • Key decision identification
    • Topic classification
    • Participant identification
    • Next steps generation
  • Fallback Mechanism: Basic keyword analysis when AI services unavailable

Data Storage Strategy

  • Local-First: All data stored in browser session state
  • Privacy by Design: No external data transmission for core functionality
  • Session Persistence: Meeting data maintained during browser session
  • Export Options: Multiple format support for data portability

Export System

  • Format Support: Markdown export with structured meeting data
  • Data Preservation: Complete meeting information including metadata
  • User Control: Local file generation for meeting records

External Dependencies

AI Models and Services

  • faster-whisper: Local speech-to-text transcription model
  • OpenAI API: Cloud AI analysis service (optional, requires API key)
  • Ollama + Gemma 3n: Local AI model integration for privacy-first analysis
  • LM Studio: Local server with OpenAI-compatible API endpoints for custom model hosting

Audio Processing

  • pydub: Audio file manipulation and format conversion
  • AudioSegment: Audio metadata extraction and processing

Web Framework

  • Streamlit: Main web application framework
  • pandas: Data manipulation for meeting records

Development Tools

  • tempfile: Temporary file handling for audio processing
  • json: Data serialization for storage and export
  • datetime: Timestamp management for meeting records

System Requirements

  • Python Environment: Python 3.11+ with Streamlit deployment
  • Audio Codecs: Support for WAV, MP3, MP4, M4A, FLAC, OGG formats
  • Model Storage: Local caching for Whisper models (auto-downloaded)
  • Browser Compatibility: Modern browser for Streamlit interface
  • Optional: Ollama for local AI processing
  • Optional: LM Studio for local server-based AI processing
  • Optional: OpenAI API key for cloud analysis

Documentation Structure

  • README.md: Comprehensive setup and usage guide with privacy modes
  • SETUP.md: Detailed installation instructions for multiple methods (uv, pip, direct)
  • replit.md: Technical architecture and project memory
  • Streamlit Config: Optimized deployment settings in .streamlit/config.toml

Recent Changes

August 2025 - Triple-Mode AI Architecture Complete

  • ✅ Created comprehensive README.md with correct Streamlit setup instructions
  • ✅ Added SETUP.md with multiple installation methods (uv, pip, direct)
  • ✅ Implemented triple-mode AI integration (OpenAI + Ollama + LM Studio)
  • ✅ Added LM Studio integration with configurable host/port/model settings
  • ✅ Enhanced CPU/non-CUDA device support for Whisper with automatic fallbacks
  • ✅ Added interactive provider selection UI with real-time connection testing
  • ✅ Improved privacy options with three levels of local processing
  • ✅ Fixed AI Configuration interface with clickable dropdown and help buttons
  • ✅ Added comprehensive documentation and setup guides for all AI providers
  • ✅ Implemented OpenAI-compatible API integration for LM Studio servers

August 2025 - Live Recording and Virtual Meeting Integration

  • ✅ Added Live Recording tab with microphone recording capabilities
  • ✅ Implemented virtual meeting platform integration guides (Zoom, Teams, Google Meet)
  • ✅ Added system audio capture instructions for local setups
  • ✅ Created comprehensive recording methods guide with pros/cons analysis
  • ✅ Added graceful fallback for web environments without audio hardware access
  • ✅ Implemented real-time recording controls with device selection
  • ✅ Added recording instructions for note-taking app methods