add AI Assistant

.gitignore

@@ -38,3 +38,14 @@ yarn-error.log*
 **/public/workbox-*
 **/public/sw.*
 **/public/fallback-*
+
+# Knowledge base databases (contains user data + embeddings)
+*.db
+*.db-journal
+rags/
+
+# Claude Code local settings
+.claude/
+
+# TypeScript build info
+*.tsbuildinfo

README.md

@@ -1,5 +1,9 @@
 <h1 align="center"><a href="https://flowoss.com">Flow - Open Source Software (OSS)</a></h1>
 
+<p align="center">
+  <a href="README.md">English</a> | <a href="README.zh-CN.md">简体中文</a>
+</p>
+
 <h2 align="center">Redefine ePub reader</h2>
 
 <p align="center">Free. Open source. Browser-based.</p>
@@ -19,9 +23,155 @@
 - Share/Download book with link
 - Data export
 - Cloud storage
+- **AI-powered features**:
+  - Knowledge Base (RAG) - Upload PDFs and build a searchable knowledge base
+  - Smart Translation - Context-aware translation with knowledge base integration
+  - Page Summarization - AI-generated summaries with knowledge base context
+  - Streaming Output - Real-time token-by-token display
+  - Configurable Context Length - Adjust how much knowledge base context to include
 
 For planed features, see our [roadmap](https://pacexy.notion.site/283696d0071c43bfb03652e8e5f47936?v=b43f4dd7a3cb4ce785d6c32b698a8ff5).
 
+## Quick Start
+
+The easiest way to get started is using our one-click startup scripts:
+
+### Windows
+
+Double-click `start.bat` or run in terminal:
+
+```bash
+start.bat
+```
+
+### Mac/Linux
+
+Open terminal and run:
+
+```bash
+chmod +x start.sh
+./start.sh
+```
+
+**Mac users**: You can also double-click `start.sh` in Finder after setting execute permissions.
+
+The script will:
+
+- ✅ Check if Node.js 18+ is installed
+- ✅ Auto-install pnpm if missing
+- ✅ Create default configuration file
+- ✅ Install dependencies
+- ✅ Start the application
+- ✅ Open browser automatically after ~5 seconds
+
+After startup, visit: **http://localhost:7127**
+
+**Note**: The browser opens after a 5-second delay to allow Next.js to compile the application. If the page shows "not accessible", wait a few more seconds and refresh.
+
+### Prerequisites
+
+- [Node.js](https://nodejs.org) (version 18 or higher)
+- [pnpm](https://pnpm.io/installation) (auto-installed by the script)
+- [Git](https://git-scm.com/downloads)
+
+## AI Features
+
+Flow integrates AI capabilities to enhance your reading experience through Retrieval-Augmented Generation (RAG).
+
+### Configuration
+
+1. Click the **AI Assistant** button (floating button on the reading page)
+2. Click the **Settings** icon to configure:
+   - **API Key**: Your OpenAI-compatible API key (e.g., OpenAI, Azure OpenAI, 阿里百炼)
+   - **API URL**: API endpoint (default: `https://api.openai.com/v1`)
+   - **Model Name**: Chat model (e.g., `gpt-4`, `qwen-plus`)
+   - **Embedding Model**: Embedding model (e.g., `text-embedding-3-small`, `text-embedding-v3`)
+   - **Context Length**: How many characters from each knowledge base entry to include (default: 1000)
+
+Configuration is saved in your browser's localStorage and never sent to any server except your configured API endpoint.
+
+### Knowledge Base
+
+The Knowledge Base feature allows you to build a local RAG system:
+
+1. **Select Storage Folder**: Click "Select Folder" to choose where to store the SQLite database
+2. **Upload PDFs**: Click "Upload PDF" and select PDF files
+3. **Automatic Processing**:
+   - PDF parsing and text extraction
+   - Chunking (splitting into ~500 character segments)
+   - Vector embedding generation
+   - Storage in local SQLite database
+4. **Enable/Disable**: Toggle knowledge bases on/off to control which ones are used for context
+
+**Performance**: Batch embedding API calls reduce processing time by 10-50x compared to single requests.
+
+### Using AI Features
+
+#### Translation
+
+**Select Text Translation**:
+
+- Select any text in the book
+- Click "Translate" in the popup menu
+- Translation appears with streaming output (token-by-token)
+- Automatically includes relevant knowledge base context
+
+**Full Page Translation**:
+
+- Click the AI Assistant button
+- Click "Translate Page"
+- Translates all paragraphs on the current page concurrently (3x-5x faster than sequential)
+- Progress bar shows real-time completion status
+- Translations are inserted below each original paragraph
+
+#### Summarization
+
+- Click the AI Assistant button
+- Click "Summarize"
+- AI generates a summary of the current page
+- Uses knowledge base context for better accuracy
+- Streaming output with "Stop" button to cancel
+- Summary appears in a modal dialog
+
+### How RAG Works
+
+1. **Query**: When you translate or summarize, the current text is embedded into a vector
+2. **Search**: Cosine similarity search finds the top 3 most relevant chunks in your knowledge base
+3. **Context Building**: Relevant chunks are formatted as reference material
+4. **Prompt Injection**: Context is prepended to your prompt
+5. **Generation**: AI model generates response with knowledge base context
+
+Example prompt structure:
+
+```
+参考以下资料：
+[参考资料1]
+{knowledge base chunk 1}
+
+[参考资料2]
+{knowledge base chunk 2}
+
+请将以下文本翻译成中文，保持原文的格式和风格：
+
+{selected text}
+```
+
+### Supported API Providers
+
+Flow works with any OpenAI-compatible API:
+
+- **OpenAI**: `https://api.openai.com/v1`
+- **Azure OpenAI**: Your Azure endpoint
+- **阿里百炼 (Alibaba Cloud)**: `https://dashscope.aliyuncs.com/compatible-mode/v1`
+- **Local Models**: Ollama, LM Studio, etc. (with OpenAI-compatible API)
+
+### Privacy & Security
+
+- **Local Storage**: Knowledge base database (SQLite) is stored locally in your browser
+- **No Server**: AI configuration and knowledge base data never leave your device
+- **Direct API Calls**: Requests go directly from your browser to your configured API endpoint
+- **No Tracking**: No analytics or telemetry for AI features
+
 ## Development
 
 ### Prerequisites