Vector database · RAG-native

Search a billion vectors before the spinner starts.

Vektor indexes your embeddings, filters on metadata, and reranks in a single hop. Ground your model on real data with recall you can measure and latency you can promise.

Create an index Read the benchmarks

Retrieval, done in one hop

A query engine, not a bucket of vectors

Hybrid search, one query

Dense vectors and keyword sparse signals fused with a tunable weight, so exact terms and semantics both land in one call.

Filter without the penalty

Metadata predicates run inside the index, not as a post-filter, so a tenant-scoped query stays as fast as an open one.

Rerank in the same hop

A cross-encoder pass sharpens the top-k before results leave the cluster. No second round trip, no glue service.

Live upserts, no reindex

Write new vectors and query them in the same second. The graph updates in place while reads stay consistent.

3.2ms

p50 query latency

0.98

Recall @ 10

10B

Vectors / cluster

60k

Queries / sec / node

Priced by the query, not the seat

Starter

One index up to a million vectors. Everything you need to ship a prototype and prove recall on your own data.

Create an index

Scale

$0.09 / M queries

Autoscaling replicas, hybrid search, and metadata filtering. Pay for what you query, nothing while idle.

Create an index

Enterprise

Let us talk

Dedicated clusters, private networking, a 99.99% SLA, and BYOC deployment inside your own cloud account.

Create an index

Point your retriever at something that keeps up.

A million vectors free, forever. Import from Pinecone, pgvector, or a parquet file in one command. No card to start.

Create an index