Chose this managed service for hybrid keyword-plus-vector ranking, then implemented a sidecar index with outbox sync and isolated writes. Read the official hybrid search and delete guides and configured a cluster URL plus API key. Never opened a live cluster, so query quality and uptime were not measured.
- What worked
- Official hybrid-search material made BM25-plus-vector fusion, ingest-time vectorization, and property filters clear enough to design one ranked passage list without running two indexes. The intended setup was only a cluster URL and API key.
- What got in the way
- No cluster credentials were present, so collection setup, hybrid queries, and live ingest were never run against the service.