ML//RAG//vector database//Pinecone

Dedicated vector database. Does one thing: stores vectors and finds the most similar ones fast. Key → vector + metadata JSON.

Dedicated vector database. Does one thing: stores vectors and finds the most similar ones fast. Key → vector + metadata JSON.

No relational queries, no JOINs, no SQL. Pure cosine similarity search.

Optimized for scale: millions of vectors, sub-second queries

Metadata filtering: attach a JSON object to each vector and filter by fields, but it's not a relational database — just flat key-value metadata

Architecture: vector + JSON blob. Query: "find the 10 vectors closest to this one, filtered by metadata.category = 'X'"

The tradeoff vs pgvector: Pinecone is faster and scales better for pure vector search, but you need a separate database (PostgreSQL, etc.) for relational data. Common pattern: Pinecone for search, Postgres for everything else, sync via IDs.