A browser-based article summarization tool that uses WebLLM to run open-source language models directly in your browser. Upload articles in bulk via CSV/XLSX and get AI-generated summaries and tags without sending your data to external servers.

• 🚀 Browser-based LLM processing - no server required
• 📊 Bulk article processing via CSV/XLSX uploads
• 🔒 Privacy-focused - all processing happens locally
• 📝 Generates concise summaries and relevant tags
• 📋 Interactive data grid for easy content management
• 💾 Export results in XLSX or CSV format

Note: the app is currently deployed at https://summarize-hub.vercel.app/

1. Clone the repository:

git clone https://github.com/scottguthart/summarize-hub.git
cd summarize-hub

2. Install dependencies:

npm install

3. Run the development server:

npm run dev

4. Open http://localhost:3000 in your browser

SummarizeHub uses @mlc-ai/web-llm to run open-source language models directly in your browser. Currently, it uses the Gemma 2B-IT model (quantized to 4-bit for efficiency) for generating summaries and tags.

Key benefits of this approach:

• Privacy: All processing happens locally in your browser
• No API Keys: No need for API keys or usage limits
• Free: No ongoing costs for API calls
• Offline Capable: Once the model is downloaded, it can work offline

• Model: gemma-2-2b-it-q4f16_1-MLC
• Size: ~1.2GB (downloaded once and cached)
• Capabilities: Instruction-tuned for better task understanding
• Format: 4-bit quantized for browser efficiency

1. Click "Download Template" to get the article template
2. Fill in your articles in the template (maximum 250 articles per batch)
3. Click "Load File" to upload your filled template
4. Click "Process Articles" to process all articles
5. Export the results using "Export XLSX" or "Export CSV"

• Browser-Based Processing:

  • Models are downloaded and stored in the browser
  • Not ideal for users with limited bandwidth or storage
  • Processing speed depends on client hardware
  • Browser storage limits may affect model caching

• Article Processing:

  • Maximum article length is 30,000 characters (~4,000 words or 10 single-spaced pages)
  • Articles are automatically chunked for efficient processing
  • Long articles are split into smaller segments at sentence boundaries
  • Each segment is summarized independently and then combined
  • Processing time increases with article length
  • Articles exceeding the length limit will be marked with an error

• Batch Processing:

  • Maximum 250 articles per batch for optimal performance
  • 10MB file size limit
  • Larger datasets should be split into multiple batches
  • Processing time varies based on article length and complexity

• Model Size:

  • Initial model download is ~1.2GB
  • Requires stable internet connection for first use
  • Model is cached in browser for subsequent use

• Support for multiple models

  • Allow switching between different models
  • Compare summaries from different models
  • Model performance metrics

• Enhanced Data Validation

  • Column data type validation
  • Required field checks
  • Custom validation rules

• Flexible File Uploads

  • Support more file formats
  • Less rigid column requirements
  • Custom column mapping

• Backend Integration

  • User authentication
  • Database storage
  • API endpoints for programmatic access

• Advanced Features

  • Concurrent summarization
  • Custom summary instructions
  • Model cache management
  • Progress persistence
  • Batch processing controls

• Server-Side Processing Option

  • Offload processing from client
  • Support for larger models
  • Better handling of large datasets

• Progressive Web App

  • Offline support
  • Background processing
  • Push notifications