# PGSync adds semantic search without data leaving your infrastructure

> Source: <https://pgsync.com/pro/>
> Published: 2026-07-26 00:30:48+00:00

# PGSync Pro[¶](#pgsync-pro)

Real-time **semantic & hybrid search** for PostgreSQL — that only re-embeds
when *meaning* changes.

Your database stays the source of truth — PGSync Pro keeps a fresh vector index
in lock-step, and only pays to embed when *meaning* actually changes. Built on
PGSync, which syncs **PostgreSQL, MySQL and MariaDB** to Elasticsearch /
OpenSearch.

See it live — no signup

Type an *idea*, not keywords. On the live demo, searching **“a rebellion
against a watchful state”** returns **Nineteen Eighty-Four** — a book that
shares none of those words. That's the difference between matching *words* and
matching *meaning*, kept in sync with the database in real time.

## What you get[¶](#what-you-get)

-
**Always fresh**

Vectors stay in sync with your database via CDC. No nightly reindex jobs, no stale search results.

-
**The change-guard**

Re-embed only when the embedded text changes. Unrelated column updates are free — and easy on your embedding bill.

-
**Semantic + hybrid**

Vector kNN fused with keyword BM25 (reciprocal rank fusion) — the best of meaning and exact match.

-
**Runs in your infra**

Local

`sentence-transformers`

embeddings mean text never leaves your environment. OpenAI / Cohere / Voyage optional. -
**Agent-ready (MCP)**

An MCP server exposes search to Claude Desktop, Claude Code, or your own agent — retrieval over live data.

-
**Preflight checks**

`pgsync-pro validate`

verifies your schema and connections before indexing — catching mistakes with actionable errors.

PGSync and django-pgsync stay free and MIT forever. Pro adds the semantic layer
on top — the same `schema.json`

and CLI, plus an `embedding`

block.

**Next:** try the live demo, watch the pipeline animated, then pick a plan.