📖 Semantic Bookmark Manager

A web application for managing bookmarks with semantic search capabilities. Traditional bookmark managers can become disorganized and difficult to navigate as they grow. Unlike a categorical approach, this tool eliminates the need to organize bookmarks manually, simplifying the user experience.

It is a simple Flask app with minimal CSS and JavaScript, quick to set up and easy to use. Features focus only on quickly adding and retrieving bookmarks.

💻 Demo

🔥 Features

• Adding one or multiple bookmarks
• Auto scraping content from the URL and generating summaries
• Semantic search using the RAG technique
• API endpoints for integration with other tools
• Desktop and mobile layout
• Data stored in a CSV file
• Support models:
  • ♊ Google Gemini
    • Embedding model: gemini-1.5-flash
    • LLM model: text-embedding-400
  • 🦙 Ollama
    • Embedding model: all-minilm
    • LLM model: llama3.1
• 🏷️ Tagging system
  • View tags under bookmark descriptions
  • Search bookmarks by tags
  • Automatically suggest tags based on bookmark content

Prerequisites

• One of the following:
  • Ollama
  • Google API credentials

💻 Setup

🐳 Docker

1. Clone the repository
2. Run docker compose up -d
3. Access the web interface at http://localhost:9696

⚒ Manual

1. Clone the repository
2. Install dependencies:

   pip install -r requirements.txt
   

3. Set up your configuration:

   • For local Ollama setup:

     • Ensure Ollama is installed and running on your machine
     • Configure Ollama to accept external connections (if the app and Ollama are running on different machines):

       export OLLAMA_HOST=0.0.0.0
       

   • For Google Gemini setup:

     • Obtain Google API credentials and set them up in the configuration

   • Start the Flask application:

     python app.py
     

   • Access the web interface at http://localhost:5000

   • Use the configuration page to set up your preferred AI service (Ollama or Google Gemini)

🔗 API Integration

The application provides API endpoints for integration with other tools. Details are below:

• POST /api/search

Request format:
{
    "query": "search query string",
    "page": 1,  // optional, default is 1
    "per_page": 10,  // optional, default is 10
    "tags": ["tag1", "tag2"]  // optional, filter by tags
}

Returns:
{
    "results": [
        {
            "link": "bookmark url",
            "summary": "bookmark summary",
            "base_url": "base url of the bookmark",
            "timestamp": "bookmark creation timestamp",
            "similarity": float,  // similarity score
            "tags": ["tag1", "tag2"]  // tags associated with the bookmark
        },
        ...
    ],
    "total_results": int,
    "page": int,
    "per_page": int,
    "total_pages": int
}

• POST /api/add

Request format:
{
    "url": "https://example.com"
}
Returns:
- 201: Bookmark added successfully
- 400: Missing url parameter or bookmark already exists
- 500: Internal server error

📝 Notes

• At least two AI models are required: one for embedding and one for language processing (LLM)
• Google Gemini is recommended as it is cloud-based and the Free Tier is more than enough to run this app (until they change the pricing 😟)
• Ollama support is available, but you need to run it on the server side for mobile devices to work (until Ollama has support for mobile devices 👍)
• When switching between models, you need to reembed all the urls. This is due to the different numpy array shape between text-embedding-004 and all-minilm.

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📜 License

This project is open-source and available under the MIT License.