LlamaIndex.TS

LlamaIndex is a data framework for your LLM application.

Use your own data with large language models (LLMs, OpenAI ChatGPT and others) in JS runtime environments with TypeScript support.

Documentation: https://ts.llamaindex.ai/

Try examples online:

What is LlamaIndex.TS?

LlamaIndex.TS aims to be a lightweight, easy to use set of libraries to help you integrate large language models into your applications with your own data.

Compatibility

Multiple JS Environment Support

LlamaIndex.TS supports multiple JS environments, including:

Node.js (18, 20, 22) ✅
Deno ✅
Bun ✅
Nitro ✅
Vercel Edge Runtime ✅ (with some limitations)
Cloudflare Workers ✅ (with some limitations)

For now, browser support is limited due to the lack of support for AsyncLocalStorage-like APIs

Supported LLMs:

OpenAI LLms
Anthropic LLms
Groq LLMs
Llama2, Llama3, Llama3.1 LLMs
MistralAI LLMs
Fireworks LLMs
DeepSeek LLMs
ReplicateAI LLMs
TogetherAI LLMs
HuggingFace LLms
DeepInfra LLMs
Gemini LLMs

Getting started

npm install llamaindex
pnpm install llamaindex
yarn add llamaindex

Setup TypeScript

{
  compilerOptions: {
    // ⬇️ add this line to your tsconfig.json
    moduleResolution: "bundler", // or "node16"
  },
}

Why?

We are shipping both ESM and CJS module, and compatible with Vercel Edge, Cloudflare Workers, and other serverless platforms.

So we are using conditional exports to support all environments.

This is a kind of modern way of shipping packages, but might cause TypeScript type check to fail because of legacy module resolution.

Imaging you put output file into /dist/openai.js but you are importing llamaindex/openai in your code, and set package.json like this:

{
  "exports": {
    "./openai": "./dist/openai.js"
  }
}

In old module resolution, TypeScript will not be able to find the module because it is not follow the file structure, even you run node index.js successfully. (on Node.js >=16)

See more about moduleResolution or TypeScript 5.0 blog.

Node.js

import fs from "node:fs/promises";
import { Document, VectorStoreIndex } from "llamaindex";

async function main() {
  // Load essay from abramov.txt in Node
  const essay = await fs.readFile(
    "node_modules/llamaindex/examples/abramov.txt",
    "utf-8",
  );

  // Create Document object with essay
  const document = new Document({ text: essay });

  // Split text and create embeddings. Store them in a VectorStoreIndex
  const index = await VectorStoreIndex.fromDocuments([document]);

  // Query the index
  const queryEngine = index.asQueryEngine();
  const response = await queryEngine.query({
    query: "What did the author do in college?",
  });

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

main();

# `pnpm install tsx` before running the script
node --import tsx ./main.ts

Next.js

You will need to add a llamaindex plugin to your Next.js project.

// next.config.js
const withLlamaIndex = require("llamaindex/next");

module.exports = withLlamaIndex({
  // your next.js config
});

React Server Actions

You can combine ai with llamaindex in Next.js, Waku or Redwood.js with RSC (React Server Components).

"use client";
import { chatWithAgent } from "@/actions";
import type { JSX } from "react";
import { useActionState } from "react";

export default function Home() {
  const [ui, action] = useActionState<JSX.Element | null>(async () => {
    return chatWithAgent("hello!", []);
  }, null);
  return (
    <main>
      {ui}
      <form action={action}>
        <button>Chat</button>
      </form>
    </main>
  );
}

// src/actions/index.ts
"use server";
import { createStreamableUI } from "ai/rsc";
import { OpenAIAgent } from "llamaindex";
import type { ChatMessage } from "llamaindex/llm/types";

export async function chatWithAgent(
  question: string,
  prevMessages: ChatMessage[] = [],
) {
  const agent = new OpenAIAgent({
    tools: [
      // ... adding your tools here
    ],
  });
  const responseStream = await agent.chat(
    {
      message: question,
      chatHistory: prevMessages,
    },
    true,
  );
  const uiStream = createStreamableUI(<div>loading...</div>);
  responseStream
    .pipeTo(
      new WritableStream({
        start: () => {
          uiStream.update("response:");
        },
        write: async (message) => {
          uiStream.append(message.response.delta);
        },
      }),
    )
    .catch(console.error);
  return uiStream.value;
}

Cloudflare Workers

Tip

Some modules are not supported in Cloudflare Workers which require Node.js APIs.

// add `OPENAI_API_KEY` to the `.dev.vars` file
interface Env {
  OPENAI_API_KEY: string;
}

export default {
  async fetch(
    request: Request,
    env: Env,
    ctx: ExecutionContext,
  ): Promise<Response> {
    const { OpenAIAgent, OpenAI } = await import("@llamaindex/openai");
    const text = await request.text();
    const agent = new OpenAIAgent({
      llm: new OpenAI({
        apiKey: env.OPENAI_API_KEY,
      }),
      tools: [],
    });
    const responseStream = await agent.chat({
      stream: true,
      message: text,
    });
    const textEncoder = new TextEncoder();
    const response = responseStream.pipeThrough<Uint8Array>(
      new TransformStream({
        transform: (chunk, controller) => {
          controller.enqueue(textEncoder.encode(chunk.delta));
        },
      }),
    );
    return new Response(response);
  },
};

Vite

We have some wasm dependencies for better performance. You can use vite-plugin-wasm to load them.

import wasm from "vite-plugin-wasm";

export default {
  plugins: [wasm()],
  ssr: {
    external: ["tiktoken"],
  },
};

Tips when using in non-Node.js environments

When you are importing llamaindex in a non-Node.js environment(such as Vercel Edge, Cloudflare Workers, etc.) Some classes are not exported from top-level entry file.

The reason is that some classes are only compatible with Node.js runtime,(e.g. PDFReader) which uses Node.js specific APIs(like fs, child_process, crypto).

If you need any of those classes, you have to import them instead directly though their file path in the package. Here's an example for importing the PineconeVectorStore class:

import { PineconeVectorStore } from "llamaindex/storage/vectorStore/PineconeVectorStore";

As the PDFReader is not working with the Edge runtime, here's how to use the SimpleDirectoryReader with the LlamaParseReader to load PDFs:

import { SimpleDirectoryReader } from "llamaindex/readers/SimpleDirectoryReader";
import { LlamaParseReader } from "llamaindex/readers/LlamaParseReader";

export const DATA_DIR = "./data";

export async function getDocuments() {
  const reader = new SimpleDirectoryReader();
  // Load PDFs using LlamaParseReader
  return await reader.loadData({
    directoryPath: DATA_DIR,
    fileExtToReader: {
      pdf: new LlamaParseReader({ resultType: "markdown" }),
    },
  });
}

Note: Reader classes have to be added explictly to the fileExtToReader map in the Edge version of the SimpleDirectoryReader.

You'll find a complete example with LlamaIndexTS here: https://github.com/run-llama/create_llama_projects/tree/main/nextjs-edge-llamaparse

Playground

Check out our NextJS playground at https://llama-playground.vercel.app/. The source is available at https://github.com/run-llama/ts-playground

Core concepts for getting started:

Document: A document represents a text file, PDF file or other contiguous piece of data.
Node: The basic data building block. Most commonly, these are parts of the document split into manageable pieces that are small enough to be fed into an embedding model and LLM.
Embedding: Embeddings are sets of floating point numbers which represent the data in a Node. By comparing the similarity of embeddings, we can derive an understanding of the similarity of two pieces of data. One use case is to compare the embedding of a question with the embeddings of our Nodes to see which Nodes may contain the data needed to answer that question. Because the default service context is OpenAI, the default embedding is OpenAIEmbedding. If using different models, say through Ollama, use this Embedding (see all here).
Indices: Indices store the Nodes and the embeddings of those nodes. QueryEngines retrieve Nodes from these Indices using embedding similarity.
QueryEngine: Query engines are what generate the query you put in and give you back the result. Query engines generally combine a pre-built prompt with selected Nodes from your Index to give the LLM the context it needs to answer your query. To build a query engine from your Index (recommended), use the asQueryEngine method on your Index. See all query engines here.
ChatEngine: A ChatEngine helps you build a chatbot that will interact with your Indices. See all chat engines here.
SimplePrompt: A simple standardized function call definition that takes in inputs and formats them in a template literal. SimplePrompts can be specialized using currying and combined using other SimplePrompt functions.

Contributing:

Please see our contributing guide for more information. You are highly encouraged to contribute to LlamaIndex.TS!

Community

Please join our Discord! https://discord.com/invite/eN6D2HQ4aX

parhammmm/LlamaIndexTS