Book a demo
Solutions · AI & vectors

Search by meaning.
Inside Postgres.

With pgvector, Postgres stores embeddings next to your rows and finds the closest ones in milliseconds. Semantic search, RAG and recommendations, with the same database, backups and security you already trust.
Book a demoRead the RAG guide
1database, not two
< 10 mstop-k on 10M vectors
SQLfilters + vectors
something comfy for a marathon
runninghikingformalbeachCloudStride Running ShoeMarathon Glide 3Tempo Racer
CloudStride Running Shoe0.98$129
Marathon Glide 30.93$149
Tempo Racer0.92$89
-- "something comfy for a marathon" → embedding $1
SELECT name, 1 - (embedding <=> $1) AS similarity
FROM products
ORDER BY embedding <=> $1 LIMIT 3;   -- 4 ms · HNSW
How semantic search works

From rows
to meaning.

  1. 1Your rowsProducts, tickets, documents: the data you already keep in Postgres.
  2. 2Become vectorsAn embedding turns each row into a point. Similar meaning, nearby points.
  3. 3Meaning clustersRunning shoes end up near trainers, not near the rain jacket.
  4. 4One query, nearest rowsA single SQL query finds the closest matches, with your filters applied.
productstrail running shoemarathon trainerrain jacketwaterproof shellcoffee grinderespresso beansrunningraincoffee“shoes for a marathon”ORDER BY embedding <=> $query LIMIT 2;
Architecture

One less database
to keep in sync.

A separate vector database means a sync pipeline, a second security model and two bills. Postgres can do both jobs.

The usual AI stack

Two databases and a pipeline

  • ✕Embeddings drift from the source rows
  • ✕Two access-control models to audit
  • ✕Two backups, two bills, two on-calls
With keenPG

One Postgres, both jobs

  • ✔Embeddings update in the same transaction
  • ✔Row-level security applies to vectors too
  • ✔One backup, one bill, one team
Built in

Everything AI apps
actually need.

pgvector, tuned

HNSW and IVFFlat indexes, sized and tuned by us for your recall and latency targets.

Hybrid search

Combine meaning with SQL filters, joins and full-text ranking in a single query.

RAG-ready

Chunk, embed and retrieve documents for your LLM, with sources you can cite and permissions you can enforce.

Scales with you

Read replicas for query fan-out and partitioning for billions of vectors.

Governed by default

The same encryption, audit logs and row-level security as the rest of your data.

Experts on call

Our engineers help you choose embeddings, design indexes and benchmark recall.

Other solutions
MigrationLeave proprietary databases behindHigh availabilitySurvive failures without downtimeSecurity & complianceProtected at every layerAI & vectorsYou are here

Your AI is only as good
as your data layer.

Run Postgres with pgvector on your own servers, or ask us to review your RAG architecture.

Book a demoRead the RAG guide