Vector databases & embedding tools
A vector database is specialized storage built to hold embeddings (C5) and search through them by "closeness of meaning" instead of exact keyword match. It's the practical infrastructure piece that makes RAG (C1) actually work at scale.
Popular options range from fully-managed (Pinecone) to open-source/self-hosted (Chroma, Weaviate) to "bolt-on" features inside existing databases (Supabase's pgvector, which you've already used). Choosing one is mostly about scale, hosting preference, and whether you want it bundled with a database you already run.
Advisory-only node - if a dev ever proposes "we'll build a RAG feature," ask what vector store they're planning to use and why; a vague answer is a signal of inexperience with this layer.
What is a vector database actually for?