For QdrantAI Platform

Search And Query Api

Qdrant · Qdrant

Vector Database — Qdrant

Qdrant evals — Search & Query API (relift v3 InfraRed)

About Qdrant

Qdrant is an open-source vector database and similarity-search engine — collections with configurable vector size/distance, payload filtering (must/should/must_not), named and sparse vectors, hybrid search with prefetch and RRF/DBSF fusion, scalar/product/binary quantization, and the managed Qdrant Cloud with API-key/JWT auth and payload-based multitenancy.

Employees

~80

Industry

Vector Database

Headquarters

Berlin, Germany

Website

qdrant.tech

Sample tests· showing 3 of 9

#	Input	Expected behavior	Check
01	Agent runs POST /collections/{name}/points/query with query=<vector>, limit=10 to get the 10 nearest neighbors.	Use the unified Query API with query set to the nearest-vector form and limit for top-k. Project results with with_payload as needed. The Query API is the recommended entrypoint and supersedes the dedicated /points/search for new code; results are ordered by similarity score.	Pass / FailAi Platformmedium
02	A RAG pipeline should only use chunks above a minimum similarity. Agent sets score_threshold so weak matches are dropped.	Set score_threshold to filter out results below a similarity floor; combined with limit, the result set may be smaller than limit when few points clear the bar. Calibrate the threshold per distance metric (Cosine scores differ from Dot/Euclid) and handle the empty-result case (abstain rather than a…	Pass / FailAi Platformhigh
03	Search results are used to render snippets. Agent sets with_payload=true and with_vector=false.	Return payload for rendering and omit vectors (with_vector=false) unless the client re-ranks locally — vectors are large and rarely needed downstream. Use a payload key selector when only some fields are needed. This cuts response size materially on large collections.	Pass / FailAi Platformlow
Use this eval