One open source retrieval layer across your data. Built for agents.
Built for agents, not keyword search. Every AI demo works beautifully until you connect it to real company data. Then everything breaks. So we do what the industry always does: we add more infrastructure.
Before long, the “simple AI feature” has become a distributed systems project. We think this is backwards.
Agents do not need another copy of your data. They need a reliable way to retrieve reality. That is why we are building Polygres.
Polygres is one retrieval surface for agents
Polygres brings relational, graph, vector and text retrieval together around the same operational data.
PostgreSQL remains the source of truth. Your application continues using normal database connections for reads and writes. Polygres adds graph traversal, vector similarity, full-text search, fuzzy matching and graph-plus-vector retrieval over those records.
That means the data your agent retrieves can be the same data your application just wrote.
An agent can start with a semantic match, follow the relationships around that result, apply structured filters and receive a ranked set of records with their surrounding context. Polygres exposes these retrieval patterns through our API and Python SDK.
[Sign up for a free polygres account here →]
Why we call it Elasticsearch for the AI era
A similar change in the industry is about to happen again.
It needs a layer designed for software that does not just search, but investigates, decides and acts.
That is what we believe Polygres can become.
Elasticsearch for the AI era.
One retrieval layer across your data. Built for agents.
