The type of the vector store provider (e.g., "pinecone", "weaviate", "in_memory", "lancedb"). This helps in selecting the correct implementation.
Optional connectionPostgres connection string. Required unless pool or adapter is given.
Optional poolA pool the caller owns. The store opens none of its own, and close() leaves this one open.
Optional adapterOpen PostgreSQL adapter owned by the caller; exclusive with pool/connectionString. The store opens no pool of
its own and close() leaves the adapter open. With pool or connectionString as well, initialize() throws.
Optional poolConnection pool size when the store opens its own pool.
10
Optional defaultDefault embedding dimensions for new collections.
1536
Optional similarityDefault similarity metric.
'cosine'
Optional tableTable name prefix for multi-tenancy.
''
Optional manageWhen false the store runs no DDL: no CREATE EXTENSION, no collections table, and createCollection() does
nothing. The caller's migrations make the tables from PostgresVectorStore.schemaSql, and the metric is
read from similarityMetric.
true
Optional textText search configuration of the lexical column and its queries.
'english'
Optional iterativepgvector's iterative index scan (0.8.0 and later) for query() and hybridSearch(): the HNSW scan goes on
until a filtered search has its rows. Without it pgvector filters after the scan and can return fewer than top-K.
Optional efhnsw.ef_search for a search (1 to 1000).
Optional maxhnsw.max_scan_tuples for an iterative scan.
A unique identifier for this specific provider instance (e.g., "pinecone-main-prod", "weaviate-dev-local"). This ID is used by the VectorStoreManager to retrieve this provider.
Optional customAny other custom properties or configurations specific to this provider instance not covered by standard fields.
Configuration specific to the Postgres vector store.