@ragfoundry/adapter-vector-pgvector

🗄️ pgvector vector store adapter. Store and search embeddings in Postgres you already run, via the pgvector extension.

A VectorStore adapter for RAGFoundry — typed retrieval-augmented generation for TypeScript.

💡 New to RAGFoundry? This page is for adding one specific provider to an app you're already building. Starting a project from scratch is faster with the scaffolder — it asks a few questions and wires everything (including this adapter, if you pick it) for you:

npm create @ragfoundry my-app

→ @ragfoundry/create

📦 Install

npm install @ragfoundry/core @ragfoundry/adapter-vector-pgvector

🚀 Usage

Works with both import and require — this package ships both builds.

ESM:

import { createRAG } from "@ragfoundry/core"
import "@ragfoundry/adapter-vector-pgvector"

const rag = createRAG({
  vectorStore: {
    provider: "pgvector",
    connectionString: process.env.DATABASE_URL!,
    tableName: "documents",
  },
  // ...the rest of your config
})

CommonJS:

const { createRAG } = require("@ragfoundry/core")
require("@ragfoundry/adapter-vector-pgvector")

const rag = createRAG({
  vectorStore: {
    provider: "pgvector",
    connectionString: process.env.DATABASE_URL,
    tableName: "documents",
  },
  // ...the rest of your config
})

⚠️ The import/require is what registers "pgvector" at runtime — remove it and createRAG throws Unknown adapter "pgvector", even though the provider name is still typed correctly. In TypeScript it's also what makes the vectorStore block above autocomplete and type-check.

⚙️ Options

| Option | Type | Required | Default | |---|---|---|---| | connectionString | string | yes | — | | tableName | string | yes | — |

Schema

The table must exist, with a width matching your embedding model:

CREATE EXTENSION IF NOT EXISTS vector;
CREATE TABLE documents (
  id        text PRIMARY KEY,
  embedding vector(1536),
  content   text,
  metadata  jsonb
);

Metadata filters use jsonb containment, so filter: { lang: "en" } becomes metadata @> '{"lang":"en"}'. Upserts run in one transaction, so a partial failure leaves nothing behind. close() releases the connection pool.

🔗 See also

• @ragfoundry/core — the pipelines, types, and every export explained
• @ragfoundry/create — npm create @ragfoundry, scaffold a whole project

📄 License

MIT