Search and filters

import { vectorFilter } from '@anvia/core/vector-store'

const results = await store.index(embeddings).search({
  query: 'reset a password',
  topK: 5,
  filter: vectorFilter.and(
    vectorFilter.eq('tenantId', 'acme'),
    vectorFilter.gt('revision', 3),
  ),
})

filterToChromaWhere maps eq, gt, lt, and, and or to Chroma $eq, $gt, $lt, $and, and $or expressions. Metadata must use values accepted by both Anvia's vector metadata contract and Chroma.

searchIds() avoids returning reconstructed documents. asTool() exposes the same index to an agent. Filters narrow retrieval; authorize tenant and document access separately.