⚡ In-Engine Hybrid Search for Drizzle ORM & Postgres (3.2x faster than Elasticsearch, 0 API tokens)

[am-mar7]

Hey Drizzle community! 👋

If you use Drizzle with PostgreSQL, you know the pain of having to spin up and maintain an external Elasticsearch or OpenSearch cluster just to provide hybrid search (vector embeddings + keyword BM25).

We open-sourced PolyVector: an in-engine hybrid search engine that runs 100% inside PostgreSQL and works natively with Drizzle ORM via sql raw queries.

💡 Why PolyVector + Drizzle?

• Zero Sync Pipelines: No Kafka or Debezium workers needed. Your data and vector graphs live in Postgres with full ACID consistency.
• 100% Local CPU ONNX: Local embeddings via ONNX Runtime (bge-small-en-v1.5). Zero OpenAI API keys or token costs.
• Native Drizzle Support: PolyVector compiles parameterized ANSI SQL that executes directly via db.execute(sql.raw(...)).

import { drizzle } from "drizzle-orm/node-postgres";
import { sql } from "drizzle-orm";
import { PolyVectorIndex } from "polyvector-search";

const db = drizzle(process.env.DATABASE_URL!);
const index = new PolyVectorIndex({
  table: "products",
  idColumn: "id",
  embeddingsTable: "product_embeddings",
  fields: {
    name: { weight: 0.50, type: "vector" },
    description: { weight: 0.20, type: "vector" },
    fts_doc: { weight: 0.30, type: "bm25" }
  },
  distinctOn: "name"
});

const { query, params } = await index.compileQuery("trail runner", { limit: 10 });
const results = await db.execute(sql.raw(query, params));

📊 100k Benchmark vs Elasticsearch 8.x:

• 40ms vs 132ms p50 latency (3.2× faster)
• ~350MB vs 2.5GB–4GB RAM (85% less memory)

Package: npm install polyvector-search
GitHub: https://github.com/am-mar7/polyvector