Polygres

Polygres is a PostgreSQL platform for agent workflows with graph traversal, vector search, and hybrid retrieval in one system. Self-hosted or managed cloud.

AI Agent Development AI Developer Tools AI Knowledge Graph

Visit Website

Overview

Polygres is a PostgreSQL-based database platform aimed at agent workflows that need relational data, graph traversal, and vector search in one place. The product positioning centers on turning rows, relationships, and embeddings into retrievable working memory for AI agents.

The site describes managed PostgreSQL hosting alongside an open-source SDK and retrieval APIs. Its core idea is to let agents query live database tables directly, fuse different retrieval signals into one ranked answer, and return token-ready context without requiring a separate vector store or sync layer.

Core capabilities

Native graph traversal

Polygres runs graph retrieval over existing foreign-key relationships, with multi-hop traversal, shortest paths, and neighborhood search over tables already in Postgres.

Hybrid search across multiple signals

The platform combines dense HNSW vector search with sparse, full-text, filtered, quantized, and late-interaction search methods, then merges the results in a single fusion pass.

Token-ready context assembly

Retrieval returns a ranked context block that includes semantic matches, graph hops, and scalar filters, so an agent can consume one result instead of stitching together several systems.

Built on Postgres, not a sidecar

The site describes the system as working directly against live tables, without a separate vector store, sync layer, or glue code.

Interactive retrieval demos

A live playground lets visitors inspect how semantic, lexical, and typed graph signals are fused for user-memory and Wikipedia-style queries.

Where Polygres fits

Grounding agent answers in live database context
Use Polygres when an agent needs to answer a question by combining relational records, graph hops, and vector similarity in one retrieval step, such as tracing a failed order through payments and disputes.
Retrieving across linked memories and records
Use the platform for connected-memory demos or internal assistants where semantic search alone is not enough and relationships between records matter to the result.
Building hybrid retrieval workflows
Apply it to applications that need one ranked answer from several retrieval modes instead of separate passes for graph search, keyword search, and embedding lookup.
Evaluating retrieval behavior with demos
Use the public playgrounds to inspect how the system behaves on sample datasets before wiring it into a production agent or schema.

Pros and Cons

Pros

Combines relational, graph, and vector retrieval in a single database-oriented workflow.
Supports live queries against existing Postgres tables, which reduces the need for separate retrieval infrastructure.
Returns ranked context blocks that are designed to be consumed directly by agents.
Offers both self-hosted and managed cloud options, according to the pricing page.
Provides public demos that show how hybrid retrieval behaves on user-memory and Wikipedia datasets.

Cons

Several setup and onboarding details are not documented in the collected page text, including specific deployment steps and SDK/client-library coverage.
The public demo pages note rate limiting, so interactive examples may respond more slowly than the underlying product.

FAQ

What is Polygres used for?

Polygres combines PostgreSQL hosting with native graph traversal, HNSW vector search, and hybrid retrieval APIs. The site positions it for agent workflows that need relational, graph, and semantic retrieval in one query.

Can I self-host Polygres or use a managed service?

The pricing page shows a self-hosted free tier and managed cloud plans. The homepage also says you can deploy a managed instance in seconds or point the open-source SDK at an existing schema.

Does Polygres have a demo?

The public demo pages show live retrieval playgrounds for user-memory search and Wikipedia search. They are intended to demonstrate how Polygres fuses semantic, lexical, and graph signals into one ranked result set.

What is Polygres Pocket?

The Pocket page describes a separate CLI package that folds codebases or files into a queryable memory layer for coding agents. It is presented as coming soon, with support for tools like Claude Code, Codex, and Cursor mentioned on the page.

Quick Facts

Category: PostgreSQL / AI retrieval platform
Primary use: Agent memory and hybrid retrieval over relational data, relationships, and embeddings
Deployment options: Self-hosted on your own Postgres instance or managed cloud hosting
Pricing: Free self-hosted tier; paid managed plans are listed on the pricing page
Related product: Polygres Pocket, a CLI-focused memory layer for coding agents
Website: polygres.com