In-Memory RAG Chatbot in n8n with OpenAI Embeddings & Agent

Build an In-Memory RAG Chatbot in n8n that answers only from your ingested document

This n8n workflow ingests a document into an in-memory vector store using OpenAI embeddings, then launches a chat experience where an OpenAI-powered RAG agent retrieves relevant chunks and answers questions based strictly on that content.

What this workflow does

• Runs an ingestion flow on demand via a Manual Trigger.
• Downloads a demo Markdown document from GitHub using an HTTP Request step.
• Splits the document into chunks, generates OpenAI embeddings, and stores vectors in an in-memory LangChain vector store.
• Exposes a chat endpoint that receives user questions.
• Uses a retrieval tool to search the in-memory store for the top relevant chunks and feeds that context to the chat model.
• Returns grounded answers based on retrieved context—and if no relevant context is found, it responds that the information is missing.

Use cases

• Support chat for a specific knowledge base: ingest internal docs once, then answer questions using only that content.
• Demo-ready RAG for SaaS operators: validate a RAG approach quickly by swapping the demo GitHub document URL for your own source.
• Automation engineering workflows: prototype retrieval-augmented responses in n8n with a clear, predictable “ingest then chat” flow.

Technical details

• OpenAI credentials required for both embedding generation and the chat model.
• n8n nodes: HTTP Request, Manual Trigger, n8nn8n-nodes-langchainagent, n8nn8n-nodes-langchainchat trigger, and n8nn8n-nodes-langchainlm chat open ai (plus supporting node types like Sticky Note).
• Conversation memory is used alongside a retrieval tool to power RAG responses from the in-memory vector store.

Setup tip: run ingestion once to populate the in-memory store, then start asking questions through the chat endpoint. To use your own content, replace the demo Markdown URL in the HTTP Request step with your document endpoint.