WhatsApp Agent with n8n, GDrive and Pinecone
A RAG-powered WhatsApp agent that lets a non-technical team query their live Google Drive, accurately, at volume.
The question I was chasing
Could a non-technical team query their own live data just by messaging WhatsApp, and trust the answers?
An independent marketing agency wanted accurate AI retrieval over their Google Drive, but management aren't going to learn a new tool. The question was whether the interface could be the one they already live in, WhatsApp, with the hard part, keeping answers accurate as the data grows, hidden behind it.
Why it exists
This was built for a real client: an independent marketing agency that wanted a bespoke solution for more accurate AI data retrieval. Their data lived in Google Drive and kept changing, and stuffing documents into a prompt falls apart fast: too much data, stale answers, blown context windows. They needed something that stayed current and stayed correct.
The constraints
Answers had to be accurate at high data volumes, which is exactly where naive context-stuffing breaks. The data was live (files change in Drive constantly) so the system couldn't run off a stale snapshot. And it all had to reach people through WhatsApp, because the real win was zero new software for the team to adopt.
The decisions that mattered
Retrieval-augmented generation over context-stuffing: the only approach that holds up as the data grows past what fits in a prompt.
n8n watching Google Drive in real time and pushing changes into a Pinecone vector database, so the index reflects the current state of the Drive rather than a one-off import.
WhatsApp as the front door, because an interface the team already trusts beats a better one they'd have to be talked into using.
What it is
A proof-of-concept RAG agent, built for an independent marketing agency, that lets users chat with and query data from a Google Drive over WhatsApp. Behind the scenes, n8n monitors Drive changes in real time and pushes updates to a Pinecone vector database, so an OpenAI GPT can deliver fast, accurate summaries to management teams even at high data volumes. There's a walkthrough above.
Where it landed
A working proof-of-concept that did the job it was scoped for: accurate retrieval over a live, growing Drive, reachable from WhatsApp. It's also the clearest example I have of building to a real client's constraints rather than my own preferences: the interesting engineering sat in keeping the index live and the answers trustworthy, not in the chat.
Part of the Rolling Waves work archive.