RAG

RAG isn't a model, it's an architecture — retrieve, augment, generate. The choice is between frameworks that orchestrate the retrieval and the vector stores underneath.

Includes RAG frameworks (LlamaIndex, LangChain), managed vector databases (Pinecone), open-source vector stores (Weaviate, Chroma, Vespa), and hybrid-search engines.

Pick LlamaIndex when retrieval quality is the bottleneck. Pick Pinecone for zero-ops production. Pick Weaviate or Chroma for self-hosted or budget-conscious. Pick Vespa at scale beyond a few million docs.