Codexify

Fast, native semantic search for your local documents.
About · Features · Installation · Development

About

Codexify is a desktop application that brings powerful semantic search to your local documents. While there are many document search tools available, Codexify differentiates itself by being:

• Fast: Native performance with optimized vector operations
• Private: All processing happens locally on your machine
• Smart: Uses modern AI embeddings for semantic understanding
• Simple: Clean, intuitive interface for managing document collections

Features

• 📚 Document Support

  • PDF files
  • Word documents (.docx)
  • Text files
  • HTML files
  • EPUB books

• 🔍 Smart Search

  • Semantic understanding of queries
  • Results ranked by relevance
  • Cross-collection search
  • Highlighted matches

• 📁 Collection Management

  • Create document collections
  • Drag-and-drop file import
  • Batch operations
  • Collection filtering

• 🎨 Modern UI

  • Clean, minimal interface
  • Dark/light mode support
  • Responsive design
  • Native platform feel

Installation

# Clone the repository
git clone https://github.com/sidmohan0/codexify.git

# Enter the project directory
cd codexify/electron

# Install dependencies
npm install

# Start the application
npm start

Development

Codexify is built with:

• Electron for cross-platform desktop support
• Transformers.js for AI embeddings
• SQLite for document storage
• React + shadcn/ui for the interface

Project Structure

codexify/electron/
├── src/
│   ├── components/     # UI components
│   ├── lib/           # Utility functions
│   ├── model.js       # AI model handling
│   ├── db.js         # Database operations
│   └── index.js      # Main electron process
├── assets/           # Images and static files
└── package.json

See DEV_NOTES.md for detailed development information and roadmap.

Technical Architecture

┌──────────────────────────────────────┐
│            Electron App              │
├──────────────────────────────────────┤
│ ┌────────────┐       ┌────────────┐  │
│ │   React    │  IPC  │   Main     │  │
│ │  Frontend  │<─────>│  Process   │  │
│ │            │       │            │  │
│ └────────────┘       └─────┬──────┘  │
│        ▲                   │         │
│        │                   ▼         │
│ ┌──────┴──────┐    ┌──────────────┐ │
│ │  shadcn/ui  │    │ Transformers  │ │
│ └─────────────┘    │    Pipeline   │ │
│                    └───────┬──────┘ │
│                            │        │
│                    ┌───────▼──────┐ │
│                    │   SQLite DB  │ │
│                    │  (Vector DB) │ │
│                    └─────────────┘ │
└──────────────────────────────────────┘

AI Models & Processing Pipeline

1. Document Processing

   • Text Extraction: Native parsers for PDF (pdf-parse), DOCX (mammoth), HTML, EPUB
   • Chunking: Custom text segmentation with overlap for context preservation

2. Embedding Generation

   • Model: Xenova/all-MiniLM-L6-v2 (384-dimensional embeddings)
   • Framework: Transformers.js for local inference
   • Optimization: Quantized model for faster processing

3. Vector Search

   • Storage: SQLite with BLOB storage for embeddings
   • Similarity: Cosine similarity with L2 normalization
   • Ranking: Score-based with semantic highlighting

Data Flow

1. Document Ingestion

   Document → Text Extraction → Chunking → Embedding → SQLite Storage
   

2. Search Process

   Query → Embedding → Vector Similarity → Result Ranking → UI Display
   

Performance Characteristics

• Embedding Generation: ~100ms per text chunk
• Vector Search: Sub-second for collections < 10k documents
• Memory Usage: ~200MB baseline, scales with collection size
• Storage: ~1.5KB per text chunk (384-dim float32 vector)

Contributing

Contributions are welcome! Please read the Contributing Guide for details on the code of conduct and the process for submitting pull requests.

License

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