LlamaIndex.TS Guide

LlamaIndex is the best framework for RAG (Retrieval Augmented Generation). It specializes in connecting LLMs to your private data.

Core Concepts

1. Loading: Getting data from files (PDF, API, Notion) into documents.
2. Indexing: Splitting documents into chunks and storing them as vectors.
3. Querying: Finding relevant chunks and asking the LLM to answer based on them.

Basic RAG Pipeline

Here is how to build a "Chat with your Data" script in Node.js.

import { Document, VectorStoreIndex, OpenAI } from "llamaindex";

async function main() {
  // 1. Create a Document
  const document = new Document({ text: "Learn-AI is a platform for frontend engineers." });

  // 2. Create an Index (chunking + embedding happens here)
  const index = await VectorStoreIndex.fromDocuments([document]);

  // 3. Create a Query Engine
  const queryEngine = index.asQueryEngine();

  // 4. Ask a question
  const response = await queryEngine.query({
    query: "What is Learn-AI?",
  });

  console.log(response.toString());
}

Production RAG: Vector Stores

For production, you can't re-index every time. Use a Vector Database like Pinecone or Supabase (pgvector).

import { PineconeVectorStore } from "llamaindex/vector-store/PineconeVectorStore";

// Connect to Pinecone
const vectorStore = new PineconeVectorStore();

// Create index from existing store
const index = await VectorStoreIndex.fromVectorStore(vectorStore);

Data Loaders (LlamaHub)

LlamaIndex has hundreds of loaders.

import { PDFReader } from "llamaindex/readers/PDFReader";

const reader = new PDFReader();
const docs = await reader.loadData("path/to/my-manual.pdf");

Advanced Querying

• Sub-Question Query Engine: Breaks a complex question into smaller ones.
• Hybrid Search: Combines keyword search with vector search.

Integration with Next.js

Like LangChain, you can stream the output to the frontend.

// app/api/chat/route.ts
import { LlamaIndexAdapter } from 'ai';

export async function POST(req: Request) {
  const { messages } = await req.json();
  const lastMessage = messages[messages.length - 1];

  const index = await getMyIndex(); // Your logic to get index
  const queryEngine = index.asQueryEngine();

  const stream = await queryEngine.query({
    query: lastMessage.content,
    stream: true,
  });

  return LlamaIndexAdapter.toDataStreamResponse(stream);
}

Next Steps

• Build a RAG Search Project.