Retrieval engine
Vector CoreMillisecond recall across billions of vectors.
Vector Core is a purpose-built hybrid search engine combining dense vectors, sparse keywords and metadata filters into a single sub-40ms query — at any scale, on hardware you already own.
<40ms
p99 latency
10B+
vectors / cluster
96%
recall@10
5x
cheaper than hosted DBs
What makes Vector Core a beast
Hybrid dense + sparse
Blend semantic vectors with BM25-style keyword signals so exact terms and fuzzy meaning both rank correctly.
Metadata pre-filtering
Filter by tenant, ACL, date or any field before the ANN search runs — secure, fast and exact.
Quantization that keeps recall
Beast-tuned product quantization shrinks memory 8x while holding 96% recall, slashing infrastructure cost.
Live, mutable indexes
Insert, update and delete with no rebuild. The index is always queryable, even mid-write.
Technical specs
- Query latency
- <40ms p99 at 1B vectors
- Index types
- HNSW, IVF-PQ, hybrid
- Consistency
- Read-your-writes
- Deployment
- Cloud, VPC, on-prem