Search and filters

Dense search accepts the standard request. Hybrid search adds index configuration:

const index = store.index({
  dense,
  sparse,
  fusion: 'rrf',
  prefetchLimit: 40,
})

const results = await index.search({
  query: 'reset a password',
  topK: 5,
  threshold: 0.7,
  filter: vectorFilter.eq('tenantId', 'acme'),
})

filterToQdrantFilter translates eq, gt, lt, and, and or into Qdrant must, should, match, and range clauses.

For dense search, threshold is sent to Qdrant as score_threshold and is also enforced while normalizing results. For hybrid search it applies to the final fused score rather than either retriever's similarity scale.

Hybrid prefetchLimit controls how many candidates each dense and sparse branch contributes before fusion; final topK controls returned logical results. Tune both on evaluation data. Filters still require independent authorization.