// the find
mayooear/langchain-supabase-website-chatbot
Build a chatgpt chatbot for your website using LangChain, Supabase, Typescript, Openai, and Next.js.
A Next.js starter for building a RAG chatbot over your own website content — scrape pages, embed with OpenAI, store vectors in Supabase pgvector, query with LangChain. Explicitly abandoned by the author and kept only for reference. If you're starting something new today, this is the wrong foundation.
The scrape-then-embed pipeline is straightforward to follow: one script ingests URLs, another runs the chat API. Supabase pgvector is a sensible choice — you get embeddings without a separate vector DB. The custom_web_loader.ts approach with Cheerio gives you direct control over what gets indexed rather than dumping raw HTML.
Archived and unmaintained — LangChain's API has changed significantly since this was written and the version pinned in package.json will have CVEs by now. The config/notionurls.ts file name leaking into a 'website chatbot' template suggests this was repurposed from a Notion-specific repo, not designed from scratch. No chunking strategy is documented — large pages will hit token limits silently. There's no authentication on the chat API route, so anyone can burn your OpenAI quota.