Obsidian RAG Plugin

A simple Retrieval-Augmented Generation (RAG) plugin for Obsidian that allows you to query your vault using Google's Gemini AI.

Features

• Real-time monitoring of markdown files in your vault
• Automatic embedding generation and updates
• Semantic search using ChromaDB
• Conversation history support
• Markdown-aware text chunking
• File change detection (create, modify, delete)

Setting Up

1. Clone the repository into your Obsidian vault's plugins directory:

   cd YOUR_VAULT_PATH/.obsidian/plugins
   git clone [repository-url] rag-plugin

2. Set up the Python environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install required Python packages:

   pip install flask chromadb google-generativeai python-dotenv watchdog

4. Configure the environment:

   • Add your Google AI API key:

     # Configure Google AI
        genai.configure(api_key="")
     

   • Update ROOT_DIR in the code to point to your Obsidian vault path

   • Here:
     ROOT_DIR = r"C:\Users\SREEHARI\Documents\Obsidian Vault"
     

5. Start the Python backend server:

   python new.py

6. Install Node.js dependencies and start the frontend:

   npm install
   npm run dev

7. Enable the plugin in Obsidian:

   • Go to Settings → Community Plugins
   • Enable the plugin named "sample pluggin"

Quick glance at Technical Details

• Database: Uses ChromaDB for vector storage (in-memory mode)
• Embedding Strategy:
  • Chunks markdown files by sections and paragraphs
  • Maximum chunk size: 1000 characters
  • Maintains document references and metadata
• AI Model: Uses Gemini 1.5 Flash with configured parameters:
  • Temperature: 0.7
  • Top-p: 0.95
  • Top-k: 64
  • Max output tokens: 1000
• File Processing:
  • Monitors .md files only
  • Ignores hidden directories
  • Supports real-time updates
• API Endpoint:
  • POST /arraysum for querying the knowledge base ( "arraysum" doesn't make sense, i will change this eventually )
  • Maintains conversation history (last 5 exchanges)