gcloud vector-search data-object search accepts dense recall targets

Introduced 2026-09-29 · score 0.93

HOW TO USE IT

For a dense vector search that needs a query-time recall target, pass `--dense-scann-target-recall=<value>` to the search command; the value must be between 0 and 1 inclusive.

WHY THIS HELPS

Set a query-time target recall for dense index search.

REQUIREMENTS AND LIMITATIONS

• --dense-scann-target-recall=<TARGET_RECALL>: Query-time target recall for dense ScaNN search; value is in [0, 1]. Cannot be combined with `--dense-scann-search-leaves-pct` or `--dense-scann-initial-candidate-count`.
• --collection=<COLLECTION> (required): Collection to search data objects from.
• --location=<LOCATION> (required): Location of the collection.
• --vector-search-field=<VECTOR_SEARCH_FIELD> (required): Vector field to search; required for vector search.
• --vector-from-file=<VECTOR_FROM_FILE> (required): Path to a JSON file containing the dense or sparse search vector; required for vector search.
• --use-index=<INDEX>: Index resource name or ID to use for the search.
• --top-k=<TOP_K>: Number of nearest neighbors to return; defaults to 10.
• Prerequisite: Initialize the Google Cloud CLI with authentication and a project, using `gcloud init` or `gcloud auth login` and `gcloud config set project`.
• The value must be in `[0, 1]`.
• The target recall applies to dense index search at query time.

FIRST-PARTY EVIDENCE

• https://docs.cloud.google.com/sdk/docs/release-notes#58700_2026-09-29
• https://docs.cloud.google.com/gemini-enterprise-agent-platform/build/vector-search-2/query-search/search
• https://docs.cloud.google.com/sdk/gcloud/reference/vector-search/collections/data-objects/search

Check the installed version and the source before using it. Entry JSON

Report a problem with this card