Inference

Embeddings beside the index, for prototypes.

Cloud can embed text or an image and search in one call, so a prototype does not stand up a second service.

hits = db.search_text(
    "notes",
    text="how do we rotate API keys?",
    filter={"tenant": "acme"},
    model="northline-embed-small",
)

Production systems often bring their own vectors. The field is the same either way.