Vector search

Use this page when your application already produces embeddings and needs local approximate nearest-neighbour search.

LeanCorpus stores and searches vectors. It does not create embeddings, choose a model or send document text to an external service. Generate compatible vectors in your application, keep one dimension per field, then index them as VectorField values.

HNSW retrieves an approximate candidate set and exact cosine rescoring orders the shortlist. Apply an ordinary LeanCorpus filter when results must remain within a tenant, category or permission boundary. Measure recall and latency on your own corpus before enabling quantisation or changing HNSW budgets.

See also: Vector search guide, Filtered vector search, and VectorQuery.