Vector Database
also called Vector store
Storage built for finding the nearest vectors, fast.
A datastore that indexes embeddings and answers approximate nearest-neighbour queries at scale. Examples: pgvector, Pinecone, Weaviate, Qdrant, LanceDB.
For most teams, adding a vector index to the database you already run (pgvector on Postgres) beats operating a separate system. Reach for a dedicated store when scale or filtering demands it.
The choice matters less than people expect, because most are approximate-nearest-neighbour indexes with similar trade-offs between recall, latency and memory. The decision that actually changes outcomes is the chunking and embedding strategy feeding it — a good index over badly split chunks loses to a plain one over well-prepared text.
When you already have one
A team with PostgreSQL is about to adopt a dedicated vector store.
Add an extension and keep the vectors next to the relational data. One system to back up, one place for permissions to live, and joins against existing tables for free. Adopt a separate store when scale or filtering genuinely demands it, not by default.
- paperDense passage retrieval for open-domain question answering ↗
Dual-encoder dense retrieval trained on questions and passages, and the comparison to BM25 that started the dense-versus-sparse debate.
- docsFile search ↗
A hosted retrieval tool over vector stores with both semantic and keyword search — managed RAG, as opposed to building the pipeline yourself.
Embedding · Vector
A list of numbers representing meaning, so similar things land close together.
Semantic Search
Search by meaning instead of matching words.
RAG · Retrieval-Augmented Generation
Look things up, then put what you found in the prompt.
Hybrid Search
Keyword search and vector search, combined and merged.