Qdrant Vector Search Engine
Build a Similarity Search Engine on Qdrant Cloud
Store embeddings, pick a distance metric, and get the nearest vectors back in milliseconds with filters applied during the search. Start free on Qdrant Cloud with no credit card.
TRUSTED BY ENTERPRISE TEAMS
Approximate nearest neighbor
Find the Nearest Vectors Fast With Approximate Nearest Neighbor Search
Qdrant indexes vectors in an HNSW graph, a graph-based approximate nearest neighbor index that makes large-scale search fast at a small, measurable recall cost.

Tune speed against recall per query.
Raise hnsw_ef to visit more of the graph for higher recall, or lower it for speed.
Check against exact results.
Set exact: true to run a full scan and compare the approximate results with the ground truth.
Choose what sits in memory.
Keep vectors or the index on disk for collections that outgrow RAM, and keep the hot path in memory.
Similarity search
Run Similarity Search in the Metric Your Model Expects
Qdrant supports cosine similarity, dot product, Euclidean distance, and Manhattan distance for dense vectors, set per collection.
Get cosine similarity search with normalization built in.
Qdrant normalizes each vector on write and computes cosine in two steps, which gives a higher search speed.
Cut off weak matches.
Set score_threshold to exclude every result with a score worse than the value you give.
Search in batches.
Send many queries in one request; n batched searches return the same results as n single searches.
Vector similarity search
Filter and Rank in One Vector Similarity Search
Most apps filter and rank at once. Qdrant applies payload filters while it searches the vectors, so every result is both close and in scope.
Filter inside the graph.
Index a payload field, and Qdrant adds links between points that share its values, so filtered queries keep a connected path to the nearest neighbors.
Keep recall under tight filters.
In Qdrant’s published test on one million vectors, the default query planner held 99.9% to 100% recall@10 across four single-keyword filters, down to 0.012% of points.
Results
What Teams Measured After Moving Similarity Search to Qdrant
Bazaarvoice
about 98% nearest neighbor recall with sub-100 ms queries over billions of vectors on Qdrant Hybrid Cloud.
Read the Bazaarvoice case studyHimalayas
about 30 ms median and 200 ms p95 inside Qdrant across 25M+ requests a month.
Read the Himalayas case studyPathwork
query time cut from 9 seconds to 2 seconds per query after moving retrieval to Qdrant Cloud.
Read the Pathwork case studyGetting started
From Sign-Up to Your First Similarity Query
Create a free cluster.
Sign up for Qdrant Cloud with no credit card.
Create a collection.
Set the vector size and the distance metric your embedding model expects.
Upsert and query.
Upload vectors with their payloads, then send a query vector and get the nearest matches back.
