Vector Store Comparison

Store	Local / Hosted	Setup effort	Metadata filtering	Scale ceiling	Cost model	Pick it when
Chroma	Local	Trivial — pip install	Yes, built-in	Small-to-medium	Free (your disk)	Learning, prototyping
FAISS	Local	Trivial — pip install	None built-in	Medium, bounded by RAM	Free (your disk)	Batch jobs, notebooks, single-process apps
pgvector	Local or self-hosted	Moderate — Postgres extension	Yes, full SQL	Medium-to-large	Your Postgres bill	Already running Postgres
Pinecone	Hosted	Low — managed API	Yes, built-in	Large	Per-index-hour, not per-query	Large hosted workload, no ops team

Recommendation

• Learning or a new prototype: start with Chroma — zero setup cost, easy to throw away.
• Already running Postgres: pgvector keeps vectors next to the relational data they describe, one system instead of two.
• Large-scale hosted workload with no infrastructure team: Pinecone trades cost for zero operations.

See also

• Overview — the shared VectorStore interface every row above implements.
• RAG Pipeline — where a vector store plugs into a full retrieval chain.