zenrag

A minimal RAG (Retrieval Augmented Generation) library for the Vercel AI SDK. Semantic search over documents with SQLite or in-memory storage.

Features

• Chunking — Configurable text chunking with overlap
• Vector stores — SQLite (persistent) or in-memory
• Embeddings — Uses AI SDK embed / embedMany with any compatible model
• Retrieval — Semantic search with optional caching
• AI SDK integration — createRetriever and createRagMiddleware for use with generateText / streamText

Installation

bun add zenrag @openrouter/ai-sdk-provider

Usage

import { createOpenRouter } from "@openrouter/ai-sdk-provider";
import { ZenRag } from "zenrag";

const openrouter = createOpenRouter({ apiKey: process.env.OPENROUTER_API_KEY! });
const embeddingModel = openrouter.textEmbeddingModel("openai/text-embedding-3-small");

const rag = new ZenRag({
  embeddingModel,
  filePath: "./knowledge.db",
  topK: 5,
});

await rag.addDocuments([
  "Your document content here...",
  "Another document...",
]);

const docs = await rag.retrieve("search query");

API

• ZenRag — Main RAG class
• SqliteStore / MemoryStore — Vector store implementations
• createRetriever(rag) — Returns a retriever function for AI SDK
• createRagMiddleware(rag, options) — Middleware for RAG in AI pipelines
• chunkText, createChunker — Chunking utilities

License

MIT