Knowledge base / RAG: document loading, template-based chunking, embedding indexing and hybrid (vector + keyword) retrieval with reranking. Depends on agent-start-model.