LlamaIndexTS is a deprecated, MIT-licensed TypeScript data framework published on npm as llamaindex, built to connect large language models to your own data inside server-side JavaScript runtimes such as Node.js, Deno, Bun, Nitro, Vercel Edge Runtime, and Cloudflare Workers.

What it is

LlamaIndexTS is a lightweight set of libraries for TypeScript and JavaScript developers who want an LLM to reason over private data rather than generic pre-training. It lives in the npm ecosystem and ships as the llamaindex package, with separate provider packages such as @llamaindex/openai supplying model access, file readers for ingestion, and storage adapters for vector databases. The project targets server-side execution first: the README runs from Node.js 20 and upward and also lists Deno, Bun, Nitro, Vercel Edge Runtime, and Cloudflare Workers as supported environments, while noting that browser support stays limited because browser runtimes lack AsyncLocalStorage-like APIs.

The concrete problem it solves is the glue code that otherwise sits between an application and a model provider. Without a framework of this kind, a team hand-writes document loading, chunking, embedding, retrieval, and prompt assembly for every provider and re-does that work whenever the model changes. LlamaIndexTS replaces that bespoke plumbing with one abstraction layer over OpenAI, Anthropic, Groq, MistralAI, Fireworks, DeepSeek, ReplicateAI, TogetherAI, HuggingFace, DeepInfra, Gemini, and the Llama 2, Llama 3, and Llama 3.1 family, so the surrounding application code stays stable when the model behind it does not. It is the TypeScript counterpart to the LlamaIndex work the same organisation maintains in Python, and LlamaCloud and LlamaParse users are now routed to the Python cloud documentation.

Key capabilities

• Runs across Node.js >= 20, Deno, Bun, Nitro, Vercel Edge Runtime, and Cloudflare Workers, with the two edge targets carrying documented limitations.
• Swaps LLM backends through provider packages rather than rewrites, using packages such as @llamaindex/openai alongside Anthropic, Groq, MistralAI, Fireworks, DeepSeek, ReplicateAI, TogetherAI, HuggingFace, DeepInfra, and Gemini providers.
• Handles embeddings and vector-database storage so ingested documents can be retrieved at query time.
• Covers agent and chatbot patterns, as reflected in the repository topics.
• Provides a React integration path for applications that surface LLM output in a front end.
• Connects to create-llama scaffolding for starting new projects.
• Includes a Next.js playground at https://llama-playground.vercel.app/, with source in the run-llama/ts-playground repository.

Who uses it and how

• Server-side TypeScript teams running Node.js 20 or later that need LLM features without leaving the npm toolchain.
• Edge-first deployments on Vercel Edge Runtime or Cloudflare Workers, accepting the limitations the README attaches to both targets.
• Deno and Bun users, both listed as first-class environments rather than as afterthoughts.
• Prototypers who start from the hosted Next.js playground or from the create-llama scaffolding to reach a working retrieval application quickly.
• Existing adopters who must now plan a migration away from the package, since LlamaCloud and LlamaParse traffic is directed to the Python cloud documentation.

Getting started

Install the core package with npm install llamaindex, or the pnpm and yarn equivalents, then add a provider package such as @llamaindex/openai for the model you intend to call. Setup instructions for each runtime are at https://ts.llamaindex.ai/docs/llamaindex/getting_started.

How it compares

No competing commercial products are named in the material for this entry, and no other framework is named either, so LlamaIndexTS stands alone in this registry with nothing to contrast it against directly. Its only stated relationship is to the wider LlamaIndex organisation and the Python-side LlamaCloud and LlamaParse documentation.

When to use it — and when not to

This project is deprecated and no longer maintained, so it should not be chosen for new work; the honest use case is maintaining or migrating an existing TypeScript application that already depends on it. A self-hoster still operates the whole stack: a Node.js 20 or later runtime, provider credentials for whichever model vendor is selected, and a vector store for ingested documents. The recent activity, 149 open issues, and the browser limitation caused by missing AsyncLocalStorage-like APIs all point to a codebase that will not absorb fixes, so teams needing ongoing support should look at the maintained Python-side offerings instead.

npm install llamaindex
pnpm install llamaindex
yarn add llamaindex

npm install @llamaindex/openai
pnpm install @llamaindex/openai
yarn add @llamaindex/openai