Features
Real-time web search via Serper, Tavily, or SearXNG
Web page content retrieval and parsing
MCP protocol support for AI tool integration
Configurable rate limiting and concurrency
Search caching to reduce API costs
Custom user-agent and headers
SSL/TLS verification control
JavaScript rendering support (via headless browser)
Result filtering by domain, date, and type
Structured JSON output for LLM consumption
Train and fine-tune ColBERT models
Automated hard negative mining
Process pairs, labeled pairs, and triplets
Strong defaults with full parameter customization
Modular components: DataProcessor, negative miners standalone
Embed and index documents
Retrieve documents with late-interaction scoring
Zero-shot retrieval in new domains
Built-in training data deduplication and mapping
Supports non-English languages with low data
Streamlined API: RAGTrainer and RAGPretrainedModel
File-based storage for version control of training data
Requires Python; install via pip install ragatouille
Open source under Apache-2.0 license
Requires script guard (if __name__ == '__main__')
