meezan35/next-sitecore-ai

next-sitecore-ai is an open-source npm package that adds AI features to websites built with Sitecore XM Cloud and Next.js. Sitecore XM Cloud is a commercial headless content management system (a tool where editors write content that a separate website then displays). This package acts as a layer on top of it, connecting the CMS content to AI capabilities like search, suggestions, and question-answering. The package provides three React hooks that your Next.js pages can call. The first, useContentSuggestion, sends a Sitecore content item path to an AI route and streams back suggestions for improving the content. The second, useSearchEnhance, runs a text search query through an AI ranking step so results come back ordered by semantic relevance rather than keyword matching. The third, usePersonalize, picks a content variant to show a visitor based on audience segments you define. Behind those hooks is a retrieval-augmented generation (RAG) pipeline. During setup you run an ingestion script that reads your published Sitecore pages through Experience Edge (Sitecore's delivery API), breaks the text into chunks, generates vector embeddings using the OpenAI API, and stores them in a Supabase database. After that, when a user asks a question through the chat API route, the system finds the most relevant content chunks by comparing their embeddings and passes them to the language model as context. The package also ships an MCP server, which exposes five tools that AI coding assistants like Claude or Cursor can call to fetch Sitecore content, search it, and list site pages. This lets developers describe content changes to an AI tool and have it read the real CMS data while responding. Setup requires a Sitecore XM Cloud subscription with Experience Edge access, an OpenAI API key, a Supabase project with the pgvector database extension, and Next.js 14 or later. Install the package via npm, add seven environment variables, run the ingestion script once, and add API routes to your project. The package is MIT licensed.