WebRAG

About

# WebRAG - 网页知识库助手

A browser extension that builds a local RAG (Retrieval-Augmented Generation) knowledge base from web pages, powered by Ollama for complete privacy and local processing.

## Features

- 📚 Build personal knowledge base from web pages
- 🌐 Batch import links with search and filtering
- 🔍 Semantic search powered by deep learning embeddings
- 💬 AI-powered Q&A with Markdown support
- 🔒 100% local processing - your data stays private
- 📦 Multiple knowledge bases management
- 🇨🇳 Excellent Chinese language support

## Quick Start

### Installation

1. Open Chrome and navigate to `chrome://extensions`
2. Enable "Developer mode" (toggle in top right)
3. Click "Load unpacked" and select the `rag-extension` folder
4. Pin the extension to your toolbar

### Ollama Setup

**Important:** Ollama must be configured for CORS to work with browser extensions.

Start Ollama with CORS enabled:

```bash
# macOS/Linux
OLLAMA_ORIGINS="*" ollama serve
```

Download required models:

```bash
# Chat model (required)
ollama pull qwen2.5:7b

# Embedding model (required for Chinese)
ollama pull lrs33/bce-embedding-base_v1
```


## Usage

### 1. Add Current Page

Click "Add Current Page" to save the current webpage to your knowledge base.

### 2. Batch Add Links

Click "Add Page Links" to batch import multiple pages:
- Extracts all links from current page
- Search and filter links
- Select multiple pages to add

### 3. Search Knowledge Base

Enter your question and click "Search" to find relevant content using semantic search.

### 4. AI Q&A

Click "AI Answer" to get intelligent responses based on your knowledge base. Responses support full Markdown formatting including code blocks, tables, and lists.

## Configuration

Open the extension and scroll to "LLM Config" section:

- **Ollama URL**: `http://localhost:11434` (default)
- **Chat Model**: Select your installed model (e.g., `qwen2.5:7b`)
- **Embedding Model**: Select Chinese-optimized model (e.g., `lrs33/bce-embedding-base_v1`)

The extension automatically loads available models from Ollama.

## Technical Stack

| Component | Technology |
|-----------|------------|
| **Frontend** | Vanilla JavaScript + Chrome Extension API |
| **Embedding** | Ollama Embeddings API (supports Chinese) |
| **Vector Store** | IndexedDB (browser local storage) |
| **LLM** | Ollama (local inference) |
| **Markdown** | marked.js |
| **Similarity** | Cosine similarity |

## Recommended Models

### For Chinese Content
- **Chat**: `qwen2.5:7b`, `qwen:14b`, `yi:34b`
- **Embedding**: `lrs33/bce-embedding-base_v1`, `bge-large-zh-v1.5`

### For English Content
- **Chat**: `llama3.2`, `mistral`, `phi3`
- **Embedding**: `nomic-embed-text`, `mxbai-embed-large`

## Data Privacy

✅ **100% Local Processing**
- All data stored in browser IndexedDB
- AI inference runs on your local Ollama
- No data sent to external servers
- Complete privacy and security