Introduction

Why Nuxt AI Kit?

Building RAG into your Nuxt app means wiring up content extraction, chunking, embeddings, vector storage, and LLM integration—each with their own APIs and quirks.

Nuxt AI Kit handles the full pipeline:

1. Embeddings - Extracts and chunks your markdown content at build time
2. Vector Storage - Framework-agnostic adapters for any vector DB
3. LLM Agent APIs - Provider-agnostic interfaces for any AI gateway or direct API

No vendor lock-in. Swap providers without rewriting code.

Check out the installation guide.

Features

• 🔄 Full RAG Pipeline - Content extraction → chunking → embeddings → vector storage → retrieval → LLM response
• 📝 Embeddings - Markdown tokenization with smart chunking, local-first via transformers.js or cloud providers
• 🗄️ Framework-Agnostic Vector DB - sqlite-vec, libsql, pgvector, Upstash, Cloudflare Vectorize
• 🤖 Provider-Agnostic LLM APIs - Vercel AI Gateway, Cloudflare AI Gateway, direct API, Ollama, Workers AI
• 🔗 Model Context Protocol - MCP search_pages tool for Claude and other AI agents
• ⚡ Build-time Indexing - Heavy operations at build, runtime is just vector lookups

How It Works

Build Time

1. Content extracted and converted to markdown
2. Split into chunks optimized for retrieval
3. Embeddings generated via your chosen provider
4. Vectors stored in your database

Runtime

1. Query converted to embedding
2. Vector similarity search finds relevant chunks
3. For chat: chunks become LLM context

Heavy work happens once at build. Runtime is just vector lookups.

Platform Support

Deploy anywhere Nitro runs: Node.js, Cloudflare Workers/Pages, Vercel, Netlify, AWS, Azure.

Vector storage options: sqlite-vec, libsql, pgvector, Cloudflare Vectorize, Upstash.