n8n Documentation Q&A Bot (End-to-End RAG Template)

An n8n template that walks through retrieval-augmented generation end to end. Nine nodes split into two halves. The first is a one-time indexing run: crawl every page of the official n8n documentation, split it with a recursive character text splitter, embed each chunk with OpenAI Embeddings, and store the vectors in n8n’s built-in Simple Vector Store. The second is the chat interface: an AI Agent retrieves the most relevant chunks for a question, then passes them to the model with a hard instruction to answer using only that material.

The real value is not a docs bot for n8n — it is that you can point the same skeleton at your own knowledge base. Crawl, chunk, embed, retrieve, constrain: every step is a visible, separate node, and no external database is needed. This is the clearest RAG structure in our collection.

Template page stats: author ayhamjo7, published 2025-07-22, roughly 183 views (collected 2026-07-27); credits Lucas Peyrin.

What it does

• Crawls all pages of the official n8n documentation (HTTP Request plus HTML parsing nodes)
• Splits long documents into retrieval-sized chunks with a recursive character splitter
• Generates embeddings for each chunk via OpenAI and stores them in n8n’s built-in vector store
• Has an AI Agent answer from retrieved chunks under a “use only this information” instruction, reducing invented answers
• Keeps multi-turn context with a Memory node
• Exposes a publicly accessible chat page URL through the Chat Trigger

Prerequisites

• n8n, cloud or self-hosted
• An OpenAI API key (the same credential serves both the chat model and embeddings)
• No external vector database — the template uses the built-in Simple Vector Store

How to use

1. Import the template and create a credential with your API key on any one of the OpenAI nodes.
2. Apply that credential to the other two OpenAI nodes as well (both chat model and embeddings).
3. Find the Start Indexing manual trigger at the top left and execute it. Per the template, crawling and embedding the full docs takes roughly 15–20 minutes.
4. Once indexing finishes, activate the workflow, open the Chat Trigger node at the bottom left, and copy its Public URL.
5. Open that URL in a new tab and ask something like “How does the IF node work?” or “What is a sub-workflow?”
6. To use your own knowledge base, change the crawl target and parsing rules — the rest of the chain stays as is.

Use cases

• n8n users who want to see what each RAG step actually does
• Prototyping a Q&A bot over internal docs, product manuals, or a help center
• Letting new team members self-serve documentation instead of asking around
• Teaching material for demonstrating how retrieval constraints suppress hallucination

Notes

• The built-in vector store lives in memory and is wiped when the n8n instance restarts, which means re-running the whole index. For anything long-lived, swap Simple Vector Store for Qdrant, Supabase, or similar.
• Indexing takes far longer than setup (roughly 15–20 minutes versus about 2). Leave time for the first run rather than cancelling midway.
• The template description is inconsistent on one point: the prose mentions Gemini for answering, but the actual nodes use an OpenAI chat model and OpenAI embeddings. Trust the nodes.
• Crawling a full documentation site generates a lot of requests — be considerate about request rate.
• The Chat Trigger’s public URL is genuinely public. Do not index sensitive internal material behind it.
• OpenAI is an overseas service; mainland China access typically requires an international network. Both embeddings and chat calls are usage-billed — see the official pricing page for current rates.