@ragpipe/plugin-pgvector

PostgreSQL + pgvector vector store plugin for ragpipe, powered by pg.

Install

pnpm add ragpipe @ragpipe/plugin-pgvector

Usage

import { defineConfig } from "ragpipe";
import { pgVectorStore } from "@ragpipe/plugin-pgvector";

export default defineConfig({
  // ... embedding, generation
  vectorStore: pgVectorStore({
    connectionString: process.env.DATABASE_URL ?? "",
    tableName: "documents", // default
    schema: "public", // default
    ssl: false, // default
  }),
});

API

pgVectorStore(options)

Returns a VectorStorePlugin backed by a direct PostgreSQL connection.

| Option | Type | Default | Description | |---|---|---|---| | connectionString | string | — | PostgreSQL connection string (required) | | tableName | string | "documents" | Table to store documents | | schema | string | "public" | PostgreSQL schema for the table | | ssl | boolean | false | Enables SSL with rejectUnauthorized: false |

Methods

| Method | Description | |---|---| | search(vector, topK) | Runs cosine similarity search with pgvector <=> | | upsert(source, content, vector) | Inserts with ON CONFLICT (source, content_hash) dedup | | clear() | Truncates the configured table | | disconnect() | Closes the underlying PostgreSQL pool | | isReady() | Checks whether the configured table already exists | | setup(dimensions, options?) | Creates or recreates the table and HNSW index |

Database Setup

setup() creates the schema automatically:

• CREATE EXTENSION IF NOT EXISTS vector
• CREATE TABLE IF NOT EXISTS {schema}.{table}
• CREATE INDEX IF NOT EXISTS {table}_vector_idx USING hnsw

Your database still needs the pgvector extension package installed at the server level. If the extension is unavailable or permissions are missing, setup() will fail with guidance in the error message.

Examples

Local Docker / self-hosted PostgreSQL

pgVectorStore({
  connectionString: "postgresql://ragpipe:ragpipe@localhost:5432/ragpipe",
});

AWS RDS / Cloud SQL with SSL

pgVectorStore({
  connectionString: process.env.DATABASE_URL ?? "",
  ssl: true,
});

Custom schema and table

pgVectorStore({
  connectionString: process.env.DATABASE_URL ?? "",
  schema: "rag",
  tableName: "knowledge_base",
});

Notes

• This plugin uses direct SQL instead of a hosted platform SDK.
• schema and tableName are validated before being interpolated into SQL.
• Vector values and document content are sent via parameterized queries.

License

MIT