AI Frameworks Overview

Building AI apps with raw API calls is possible, but frameworks solve common problems like streaming state management, chaining, and document retrieval.

The Big Three

Framework	Best For	Philosophy
Vercel AI SDK	Frontend/Fullstack	"UI First." Built for React/Next.js/Svelte. Handles streaming hooks and UI updates effortlessly.
LangChain.js	Backend Logic/Agents	"Chains & Agents." Massive ecosystem of integrations. Great for complex workflows independent of UI.
LlamaIndex TS	Data/RAG	"Data First." Specialized in connecting LLMs to your data (PDFs, SQL, Notion) for RAG apps.

---

Decision Matrix: What should I use?

Detailed Guides

1. Vercel AI SDK (Recommended Start)

The default choice for Next.js developers. It unifies the API layer and the UI layer.

• Key Features: useChat, useCompletion, streamText.
• Ecosystem: Works with OpenAI, Anthropic, Mistral, Google, and more.

2. LangChain.js

The Swiss Army Knife of AI.

• Key Features: Prompt Templates, Output Parsers, Chains, Agents.
• Use Case: When you need to switch models easily or build complex multi-step reasoning chains.

3. LlamaIndex.TS

The Data Framework.

• Key Features: Data Loaders (PDF, HTML), Vector Store Indexing, Query Engines.
• Use Case: Building a chatbot that "talks to your data" (RAG).

Can I mix them?

Yes! A very common stack is:

• Vercel AI SDK for the Frontend UI (streaming hooks).
• LlamaIndex or LangChain on the Backend to retrieve data and construct the prompt.

Next Steps

Start with the Vercel AI SDK Guide to build your first streaming UI.