WordPress RAG Chatbot — AI Q&A over Website Content

Overview

This project turns a WordPress website into a conversational knowledge base. Visitors ask questions in natural language and receive answers grounded in the site's own posts and pages, along with source metadata such as URL, publication date and content type.

How it works

The system is split into three cooperating n8n workflows:

1. Initial indexing — fetches all published posts and pages through the WordPress API, filters out protected or unpublished content, converts HTML to Markdown, splits text with a token splitter and stores OpenAI text-embedding-3-small embeddings in a Supabase pgvector store.
2. Incremental upsert — runs on a schedule, reads the timestamp of the last execution from PostgreSQL, and fetches only content modified since then. Changed documents have their old vectors removed and fresh embeddings inserted, keeping the index current without full re-indexing.
3. Chat — a webhook/chat trigger feeds an AI agent that uses the vector store as a retrieval tool and PostgreSQL-backed chat memory for multi-turn context, returning answers through a respond-to-webhook node so the chat can be embedded on any site.

The architecture demonstrates a complete production RAG loop — ingestion, change tracking, embedding, retrieval and conversational generation — built entirely from n8n's LangChain nodes.