Embeddings + Vector DB for RAG
Gateway in front of both the embedding model and the chat model — the full RAG loop with Pinecone, Qdrant, or pgvector, including the embedding-cost question.
💡
Tip: RAG with the gateway in front of both the embedding call and the chat call is the cleanest seam — one place for cost, one place for retries, one place to swap providers.
What this use case covers
- The full RAG loop: chunk → embed (via gateway) → store (Pinecone / Qdrant / pgvector) → retrieve → augment → chat (via gateway).
- Gateway in front of the embedding model, not just the chat model — so embedding cost and rate limits land in the same dashboards.
- The embedding-cost question: run once, cache forever (with proper version keys).
- A short comparison of Pinecone vs Qdrant vs pgvector for the use case at hand.