@farming-labs/grag

TypeScript GraphRAG primitives for apps that want graph-backed retrieval, citations, and relational storage without leaving the TypeScript ecosystem.

Install

npm install @farming-labs/grag kysely

Install the database driver you use separately:

npm install pg
npm install better-sqlite3
npm install mysql2

What You Get

• GraphRAG artifact models for documents, chunks, entities, relationships, communities, reports, and embeddings.
• Memory storage for tests and prototypes.
• SQL/Kysely storage with migrations for relational databases.
• Optional @farming-labs/orm adapter from @farming-labs/grag/orm.
• Retrieval helpers for lexical search, graph context, local search, global search, citations, and source cards.
• A CLI Studio for previewing and querying GraphRAG snapshots.

Quick Start

import {
  MemoryGraphRagStore,
  chunkDocuments,
  createGraphRagService,
  relationalRowsToDocuments,
} from "@farming-labs/grag";

const rows = [
  { id: 1, customer: "Acme", body: "The billing export failed after migration." },
  { id: 2, customer: "Globex", body: "Password reset worked." },
];

const documents = relationalRowsToDocuments({
  tableName: "support_tickets",
  rows,
  idColumn: "id",
  titleColumn: "customer",
  textColumn: "body",
});

const graph = chunkDocuments(documents);
const store = new MemoryGraphRagStore();
await store.upsertGraph(graph);

const grag = createGraphRagService({ store });
const answer = await grag.ask("What failed after the migration?");

console.log(answer.answer);
console.log(answer.citations);

Database Source Adapter

Use source.database when your app should own the SQL query and GRAG should handle row-to-document conversion, row provenance, chunking, and retrieval citations:

import { DataSourceLoader, source } from "@farming-labs/grag";

const loader = new DataSourceLoader([
  source.database({
    label: "Production support tickets",
    tableName: "support_tickets",
    loadRows: () =>
      db.selectFrom("support_tickets").selectAll().where("status", "!=", "spam").execute(),
    idColumn: "id",
    titleColumn: "subject",
    textColumn: "body",
    attributeColumns: ["customer", "priority", "status", "created_at"],
  }),
]);

const { documents, textUnits } = await loader.load();
await store.upsertGraph({ documents, textUnits });

SQL Storage

import { Kysely, PostgresDialect } from "kysely";
import { Pool } from "pg";
import {
  SqlGraphRagStore,
  applyGraphRagMigrations,
  type GraphRagSqlDatabase,
} from "@farming-labs/grag";

const db = new Kysely<GraphRagSqlDatabase>({
  dialect: new PostgresDialect({
    pool: new Pool({ connectionString: process.env.DATABASE_URL }),
  }),
});

await applyGraphRagMigrations(db, "postgres");

const store = new SqlGraphRagStore({ db });

OpenAI Helpers

import { createGraphRagService } from "@farming-labs/grag";
import { OpenAiChatModel, OpenAiEmbeddingModel } from "@farming-labs/grag/openai";

const grag = createGraphRagService({
  store,
  model: new OpenAiChatModel(),
  embeddingModel: new OpenAiEmbeddingModel(),
});

const result = await grag.ask("How does retrieval work?", {
  responseStyle: "short answer with citations",
});

CLI Studio

grag studio ./snapshot.json --port 3333 --open
grag retrieve ./snapshot.json "How does graph expansion help retrieval?"
grag repo . --ask "What does this repository do?" --snapshot repo.snapshot.json

Docs

• Getting started
• Retrieval SDK
• Storage configuration
• Service integration
• Architecture

Development

pnpm install
pnpm --filter @farming-labs/grag check
pnpm --filter @farming-labs/grag test
pnpm --filter @farming-labs/grag build

From the monorepo root, pnpm build compiles the package, CLI Studio assets, and docs app.