Vector database
A database built to store embeddings and find the nearest ones to a query vector fast. The recurring cost centre of a RAG system.
What it is
It indexes vectors so that "find the 10 closest to this one" runs in milliseconds over millions of entries, using approximate nearest-neighbour algorithms (HNSW, IVF). Options include managed services (Pinecone, Zilliz, Qdrant Cloud), the pgvector extension for Postgres, and self-hosted engines.
Why it matters
Pricing models vary wildly: some charge for stored vectors and queries (serverless), others for the RAM/compute of a running cluster. This makes head-to-head comparison hard - normalise on cost for a reference workload.
Cost & infrastructure impact
For 1M vectors and moderate query volume, expect roughly $20-75/month depending on provider and dimension. Storage scales with vector count x dimensions; query cost scales with traffic.