Create a new VectorStoreClient.
Vector store configuration.
OptionalapiKey?: stringQdrant API key.
Qdrant collection name.
Qdrant server URL.
The embedding vector dimensions.
Optionallogger: Logger<never, boolean>Optional pino logger for retry warnings.
Delete points by their IDs.
Uses a fresh QdrantClient per attempt to avoid stale keep-alive connections.
The point IDs to delete.
Ensure the collection exists with correct dimensions and Cosine distance.
Create a full-text payload index on the specified field. Idempotent — skips if the field already has a text index.
The payload field to index.
Get collection info including point count, dimensions, and payload field schema.
When Qdrant has payload indexes, uses payload_schema directly. Otherwise
samples points to discover fields and infer types.
Hybrid search combining dense vector and full-text match with RRF fusion.
The query vector.
The raw query text for full-text matching.
Maximum results to return.
Weight for text results in RRF (0–1).
Optionalfilter: Record<string, unknown>Optional Qdrant filter.
An array of search results.
Scroll through all points matching a filter.
Optionalfilter: Record<string, unknown>Optional Qdrant filter.
Page size for scrolling.
Search for similar vectors.
The query vector.
Maximum results to return.
Optionalfilter: Record<string, unknown>Optional Qdrant filter.
Optionaloffset: numberAn array of search results.
Upsert points into the collection.
Uses a fresh QdrantClient per attempt to avoid stale keep-alive connections. Between embedding calls and upserts, idle connections may be closed by the server, causing ECONNRESET on reuse.
The points to upsert.
Client wrapper for Qdrant vector store operations.
Implements the VectorStore interface for dependency inversion.