Pgvector: vector(n) columns, distance operators (<-> L2, <=> cosine, <#> inner product), HNSW vs IVFFlat indexes and tuning (m, ef_search, lists, probes)

With the extension created, you store embeddings in a column of type vector(n), where n is the number of dimensions your embedding model returns. Inserting a vector with a different length fails, which protects you from mixing models. A typical table keeps the chunk text, its metadata and its embedding together, so a single SQL query can filter and rank.