Anvia

Appearance

@anvia/qdrant ​

@anvia/qdrant supports dense and hybrid dense-plus-sparse retrieval over Qdrant collections.

Install ​

sh
pnpm add @anvia/qdrant @anvia/core @anvia/openai @qdrant/js-client-rest

The ESM package includes the Qdrant REST client and peers with @anvia/core >=0.7.1 <1.0.0.

Dense retrieval ​

ts
import { embedDocuments } from '@anvia/core/embeddings'
import { QdrantVectorStore } from '@anvia/qdrant'
import { OpenAIClient } from '@anvia/openai'

const openai = new OpenAIClient({
  apiKey: process.env.OPENAI_API_KEY,
})
const embeddings = openai.embeddingModel('text-embedding-3-small')
const sourceDocuments = [
  {
    id: 'password-reset',
    text: 'Password reset links expire after 30 minutes.',
  },
]

const documents = await embedDocuments(embeddings, sourceDocuments, {
  id: (document) => document.id,
  content: (document) => document.text,
})

const store = await QdrantVectorStore.connect({
  collectionName: 'support_docs',
  vectorSize: 1536,
  distance: 'Cosine',
})

await store.upsertDocuments(documents)

const results = await store.index(embeddings).search({
  query: 'How do I reset a password?',
  topK: 5,
})

Hybrid retrieval ​

Create the collection and index in hybrid mode, then ingest documents containing aligned dense and sparse embeddings. In this example, dense, sparse, and hybridDocuments are produced by the SDK's hybrid embedding workflow:

ts
const store = await QdrantVectorStore.connect({
  collectionName: 'support_docs_hybrid',
  vectorSize: 1536,
  hybrid: true,
})

await store.upsertDocuments(hybridDocuments)

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

Hybrid collections default to named vectors dense and sparse; fusion defaults to rrf, with dbsf also supported. Dense-only and hybrid store/index modes cannot be mixed.

Collection ownership ​

By default, the adapter reads an existing collection or creates one with the configured dimension and distance. Production deployments should create collections, payload indexes, replication, and storage settings through infrastructure automation, then use createIfMissing: false.

Keys beginning with __anvia_ are reserved in payload metadata. Keep configured dense and sparse vector names identical during creation, ingestion, and query.

Production patterns ​

  • Inject an authenticated Qdrant client with explicit endpoint and transport settings.
  • Choose dense versus hybrid before ingesting the collection.
  • Tune hybrid prefetch independently from final topK.
  • Add payload indexes for common metadata filters.
  • Treat collection or payload filters as retrieval boundaries, not application authorization.

Reference ​

Built for Anvia.