pgvector

In this catalog it is the Vector DBs option when you need to store and search embeddings for retrieval. Keep pgvector on the observe/propose side of the loop: useful signals and drafts, with human or policy authority still required before anything irreversible runs.

Documents/embeddings are upserted with metadata; queries return nearest neighbors (and filters); apps use hits for RAG or memory recall.

1. Create a collection/index with the embedding dimension you will use.
2. Upsert a labeled batch of points with metadata filters.
3. Query nearest neighbors and verify filters exclude noise.
4. Point the RAG/agent host at this collection as its retrieval backend.

Related: llamaindex · apache-age · topics/04-data

• Use when: pgvector adds vector similarity types and indexes to PostgreSQL — embeddings next to relational truth in one transactional store matches the bottleneck and you specifically need to store and search embeddings for retrieval.
• Skip when: another pack already covers the need with less ops surface, or pgvector would become a second place people retype truth.
• Prefer instead: compare LlamaIndex, Apache AGE before locking the cell.

• Ops: confirm license, hosting, and release cadence for pgvector (pgvector/pgvector) before relying on it in a live corridor.
• Integration: bad credentials, scopes, or schema drift produce empty retrievals or runaway tool/model spend without an obvious UI alarm.
• Scope: Vector DBs components support agents; they are not the company ledger or the approval authority.