Qdrant vector store

Qdrant is a dedicated vector search server, selected via VECTOR_STORE_PROVIDER=qdrant.

Enable it

VECTOR_STORE_PROVIDER=qdrant
QDRANT_URL=http://localhost:6333
QDRANT_API_KEY=your-qdrant-api-key

Install the client: pip install qdrant-client.

Configuration

Variable	Default	Description
VECTOR_STORE_PROVIDER	pgvector	Set to qdrant to use this store.
QDRANT_URL	http://localhost:6333	URL of the running Qdrant server.
QDRANT_API_KEY	(none)	Optional API key for authenticated deployments.

Notes

• Unit-tested against a mocked client, not integration-tested against a real Qdrant server. Treat first production use as unvalidated and smoke-test before relying on it.
• keyword_search is degraded: Qdrant has no BM25/ts_rank, so it falls back to a scoped fetch plus term-overlap scoring. pgvector remains the optimal backend for hybrid keyword search.
• increment_access is a read-modify-write rather than an atomic SQL update, so concurrent bumps of the same point can lose an increment.
• list ordering and offset are handled client-side: Qdrant's scroll has no numeric offset, so rows are fetched and ordered by created_at in Python.
• Point ids must be UUIDs or unsigned ints (a Qdrant constraint). The engine generates uuid4 ids, so this holds in practice; arbitrary string ids would be rejected.