During the RAG indexing phase, documents are chunked and converted into vectors via an embedding model, then stored in vector databases such as Faiss, Pinecone, and Weaviate.