ML//RAG//vector database//Pinecone

Pinecone is a dedicated vector database that does one thing: it stores vectors and finds the most similar ones fast (simpler and often faster for pure vector lookup than Weaviate, which adds hybrid search and schema-aware relationships).

Pinecone is a dedicated vector database that does one thing: it stores vectors and finds the most similar ones fast (simpler and often faster for pure vector lookup than Weaviate, which adds hybrid search and schema-aware relationships).

Each key holds a vector and a JSON blob of metadata. A query asks for the 10 vectors closest to a given one, filtered by a metadata field such as metadata.category = 'X'. There are no relational queries, no JOINs and no SQL, only cosine similarity search.

It is optimized for scale, with millions of vectors and sub-second queries.

Metadata filtering works on the JSON object attached to each vector, but that metadata is flat key-value data and Pinecone is no relational database.

The common pattern pairs Pinecone for search with a separate database (PostgreSQL or another) for everything else, the two kept in step through shared IDs. pgvector avoids the second system by doing vectors inside PostgreSQL, one database and one query; Pinecone is faster at scale.