Anvia

@anvia/qdrant API reference ​

All public symbols are exported from @anvia/qdrant.

ts
import {
  QdrantHybridVectorStore,
  QdrantTenant,
  QdrantVectorClient,
  QdrantVectorStore,
  filterToQdrantFilter,
  type QdrantClientLike,
  type QdrantDenseVectorStoreOptions,
  type QdrantDistance,
  type QdrantFusion,
  type QdrantHybridVectorStoreOptions,
  type QdrantMutationOptions,
  type QdrantVectorClientOptions,
  type QdrantVectorStoreBaseOptions,
  type QdrantVectorStoreOptions,
} from '@anvia/qdrant'

QdrantVectorClient ​

ts
class QdrantVectorClient {
  constructor(options?: QdrantVectorClientOptions)
  vectorStore<T, Metadata extends VectorMetadata = VectorMetadata>(
    options: QdrantVectorStoreOptions,
  ): QdrantVectorStore<T, Metadata> | QdrantHybridVectorStore<T, Metadata>
  close(): Promise<void>
  tenant(tenantId: string): QdrantTenant
}

QdrantVectorClientOptions accepts an injected client?: QdrantClientLike, or the official Qdrant client parameters such as url and apiKey directly.

Dense store options contain collectionName, dimensions, optional metric and denseVectorName, and optional mode: 'dense'. Hybrid options require mode: 'hybrid' and may add sparseVectorName.

Tenants ​

ts
const tenant = client.tenant(tenantId)
const scoped = tenant.vectorStore({ collectionName, dimensions })

client.tenant(tenantId) returns a QdrantTenant whose stores scope points under the sha256 digest of tenantId: physical point IDs derive from the namespace-prefixed logical ID and payloads carry a namespace marker, so upsert, search, inspect, delete, and get on a tenant store only touch its own scope. tenant.namespace exposes the derived namespace.

Stores ​

ts
await store.ensure()
await store.validate()
await store.upsert({ documents, providerOptions })
const dense = await store.search({ vector, topK, minScore, filter, providerOptions, abortSignal })
const page = await store.inspect({ limit, cursor, filter, providerOptions, abortSignal })
await store.delete({ documentIds, providerOptions })
const items = await store.get({ documentIds })

QdrantHybridVectorStore also implements:

ts
const results = await store.searchHybrid({
  vector,
  sparseVector,
  fusion: 'rrf',
  topK,
  minScore,
  filter,
})

QdrantMutationOptions supports wait, ordering, and timeout. Native distances are 'Cosine' | 'Dot' | 'Euclid'; QdrantFusion follows the core 'rrf' | 'dbsf' values. filterToQdrantFilter(filter) converts an Anvia filter for direct client calls.

Return to the package guide.

Built for Anvia.