# CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

## Project Overview
SORMARK is an AI-powered bookmark management application that analyzes and categorizes Twitter bookmarks using LLMs. It fetches bookmarks from Twitter, processes them with Google's Gemini AI to categorize content, and provides a workflow to review and organize bookmarks by category.

## Tech Stack
- **Frontend**: Waku (React framework with server components)
- **Runtime**: Bun
- **Styling**: Tailwind CSS v4
- **Language**: TypeScript with strict configuration
- **Development**: Devenv for environment management
- **LLM**: Google Gemini 2.5-flash for categorization
- **Data Storage**: JSON files and potential Obsidian integration

## Development Commands

### Getting Started
```bash
# Enter development environment
direnv allow  # or devenv shell

# Install dependencies
bun install
```

### Running the Application
```bash
# Development server
bun dev

# Build for production
bun build

# Start production server
bun start
```

### Environment Setup
The project uses devenv for environment management. Key environment variables:
- `GEMINI_API_KEY`: Google Gemini API key for LLM categorization
- `TWITTER_COKI`: Cookie used on x.com frontend for bookmark access
- `TWATTER_COKI`: Alternative environment variable name for Twitter cookie

## Project Structure
```
app/
├── src/
│   ├── pages/           # Waku pages (file-based routing)
│   │   ├── _layout.tsx  # Root layout component
│   │   ├── index.tsx    # Home page - displays all bookmarks
│   │   ├── categorize.tsx # Bookmark categorization workflow
│   │   └── about.tsx    # About page
│   ├── components/      # React components
│   │   ├── counter.tsx  # Client-side counter example
│   │   ├── header.tsx   # Site header
│   │   └── footer.tsx   # Site footer
│   ├── lib/             # Core libraries
│   │   ├── twitter-api.ts    # Twitter API integration
│   │   ├── categorization.ts # Category definitions and types
│   │   ├── llm-service.ts    # Google Gemini integration
│   │   ├── llm-prompts.ts    # LLM categorization prompts
│   │   ├── bookmark-storage.ts # Bookmark storage utilities
│   │   └── testData.json     # Test bookmark data
│   └── styles.css       # Global styles with Tailwind
├── public/              # Static assets
└── package.json         # Dependencies and scripts
```

## Core Features Implemented

### ✅ Twitter Integration
- **Bookmark Fetching**: Fetches all bookmarks from Twitter using authenticated API calls
- **Media Processing**: Extracts images and video thumbnails from bookmarks
- **Rate Limiting**: Respects Twitter API limits with 1-second delays between requests
- **Error Handling**: Comprehensive error handling for API failures

### ✅ AI Categorization
- **LLM Integration**: Uses Google Gemini 2.5-flash for intelligent content categorization
- **Image Analysis**: Analyzes bookmark images to enhance categorization accuracy
- **Custom Categories**: User-defined category system with criteria
- **Multi-category Support**: Allows bookmarks to belong to multiple categories
- **Confidence Scoring**: Provides confidence levels for categorization suggestions

### ✅ Server-Rendered Categorization Workflow
- **Progressive Processing**: Processes bookmarks one-by-one with server-side LLM calls
- **Progress Tracking**: Shows "Bookmark X of Y" with visual progress bar
- **Category Selection**: Checkbox-based interface for selecting categories
- **Save & Next**: Form-based navigation to next bookmark after categorization
- **Skip Functionality**: Allows skipping bookmarks without categorization

### ✅ Bookmark Management
- **Complete Data**: Stores tweet text, author info, media, hashtags, and URLs
- **Search & Filter**: Filter bookmarks by categories and search text
- **Remove Bookmarks**: API endpoint to remove bookmarks from Twitter
- **Export Options**: Integration with Obsidian vault for note storage

## Next Steps & Roadmap

### 🔧 Immediate Improvements
1. **Database Integration**: Replace JSON storage with SQLite/PostgreSQL for persistence
2. **Category Management**: Add UI for managing custom categories
3. **Search/Filter**: Implement advanced filtering by date, author, categories
4. **Bulk Operations**: Allow bulk categorization of similar bookmarks

### 🚀 Advanced Features
1. **Smart Suggestions**: Improve LLM prompts based on user feedback
2. **Content Analysis**: Extract and analyze linked article content
3. **Calendar Integration**: Create events from bookmarks with dates
4. **RSS Export**: Generate RSS feeds for categorized bookmarks
5. **Browser Extension**: Chrome/Firefox extension for bookmarking directly

### 🎯 User Experience
1. **Bookmark Queues**: Create queues for different processing priorities
2. **Keyboard Shortcuts**: Add keyboard navigation for faster categorization
3. **Dark Mode**: Implement dark theme support
4. **Mobile Responsive**: Optimize for mobile devices

### 📝 Data Persistence
1. **Save Configuration**: Store user category preferences
2. **Processing State**: Save categorization progress across sessions
3. **Export Formats**: Add CSV, JSON, and Markdown export options
4. **Backup/Restore**: Implement bookmark backup and restore functionality

## Environment Variables Required
```bash
# Required
GEMINI_API_KEY=your_google_gemini_api_key
TWITTER_COKI=your_twitter_authentication_cookie

# Optional
LLM_BASE_URL=custom_llm_endpoint_if_needed
LLM_API_KEY=custom_llm_api_key_if_needed
```

## Development Tips
- **Test Data**: Use `testData.json` for development without Twitter API calls
- **Cookie Refresh**: Twitter cookies expire frequently - refresh as needed
- **Rate Limits**: Be mindful of Twitter API rate limits during development
- **LLM Costs**: Monitor Gemini API usage during development