Answers From Your Documents, Not From Guesswork
Questions are embedded, matched against a Pinecone index of the business's material; the top five passages become context, in the voice and scope you set.
Knowledge Retrieval Automation
RAG Pipeline and Chatbot
An n8n workflow that turns a company's own documents into an answer engine: a retrieval-augmented chatbot for a website or an internal team, and a pipeline that keeps what it knows current the moment a file changes.
Year2026
RoleAutomation & Integration
Typen8n Automation
Built withn8n
Overview
Most of what a business knows is sitting in a folder nobody reads. This is one workflow in two halves that fixes that: an ingestion pipeline that turns a Drive folder into a searchable vector index, and a chatbot that answers out of it. Put it behind a website widget or an internal channel and people ask a question instead of hunting for the file.
It's the same workflow answering in the corner of this page. Ask it something about me and you're talking to the chatbot half of it live, not a demo of it: webhook in, rate limit, retrieval, reply out.
Features
The Knowledge Base Maintains Itself
A Drive folder is polled every minute. New files are chunked, embedded, indexed; a replacement's old vectors deleted by document type first, or both answer.
Rate Limiting That Outlives a Restart
Three windows at once: visitor per minute, conversation per hour, deployment per day. Counted in n8n's database, not memory, and stopped at the first rejection.
It Remembers the Rest of the Conversation
Memory is keyed to the session, so "what does that cost?" resolves against what was asked. Two visitors at once never see each other's thread.
One Endpoint, Any Front End
One authenticated webhook takes a message and session, returns a reply, answers 429 with Retry-After when declining. Any site, tool, or chat platform calls it.
Preview
The canvas above is the whole thing: the chat path along the top, the two ingestion paths under it. The launcher in the corner of this page is that top row running.
