Cilantrobyte.

Tech Stacks · Data

pgvector

Vector search without a second database.

Category

Data

At the studio

Since 2023

Projects shipped

1 case study

Status

Active

(01) Our take

pgvector is how we avoid standing up a second database for most AI features. For applications that already run Postgres — which is most of ours — pgvector lets us add semantic search, retrieval-augmented generation, and embedding-based features without adding Pinecone or Qdrant to the operational surface area.

The performance ceiling is lower than specialised vector databases, but we've yet to hit a client use case where the gap mattered. For scale beyond 10 million vectors with high QPS, we'd consider a dedicated store — below that, pgvector is almost always the right call.

(02) What we build with it

Typical work

  • RAG (retrieval-augmented generation) systems
  • Semantic search over documents, products, or content
  • Recommendation systems that combine text and behaviour signals
  • Duplicate detection and content clustering

(03) How we engage

Engagement shape

Usually part of an AI feature build. 4–8 weeks to production for a well-scoped RAG application running on pgvector.

Thinking about a pgvector project?