📖 Build a document QA system with Google Drive, Pinecone, and OpenAI RAG

⚡ 441 views · 📖 Internal Wiki & Knowledge Base

Description

Title

RAG AI Agent for Documents in Google Drive → Pinecone → OpenAI Chat (n8n workflow)

Short Description

This n8n workflow implements a Retrieval-Augmented Generation (RAG) pipeline + AI agent, allowing users to drop documents into a Google Drive folder and then ask questions about them via a chatbot. New files are indexed automatically to a Pinecone vector store using OpenAI embeddings; the AI agent loads relevant chunks at query time and answers using context plus memory.

Why this workflow matters / what problem it solves

• Large language models (LLMs) are powerful, but they lack up-to-date, domain-specific knowledge.
• RAG augments the LLM with relevant external documents, reducing hallucination and enabling precise answers. (Pinecone)
• This workflow automates the ingestion, embedding, storage, retrieval, and chat logic — with minimal manual work.
• It’s modular: you can swap data sources, vector DBs, or LLMs (with some adjustments).
• It leverages the built-in AI Agent node in n8n to tie all the parts together. (n8n)

How to get the required credentials

You’ll configure these under n8n → Credentials → New Credential, matching credential names referenced in your workflow nodes.

Detailed Walkthrough: How the Workflow Works

Here’s a step-by-step of what happens inside your workflow (matching your JSON):

1. Google Drive Trigger

• Watches a specified folder in Google Drive. Whenever a new file appears (fileCreated event), the workflow is triggered (polling every minute).
• You must set the folder ID (in “folderToWatch”) to the Drive folder you want to monitor.

2. Download File

• Takes the file ID from the trigger and downloads the file content (binary).

3. Indexing Path: Embeddings + Storage

(This path only runs when new files arrive)

• The file is sent to the Default Data Loader node (via the Recursive Character Text Splitter) to break it into chunks with overlap (so context is preserved).
• Each chunk is fed into Embeddings OpenAI to convert text into embedding vectors.
• Then Pinecone Vector Store (insert mode) ingests the vector + text metadata into your Pinecone index.
• This ensures your vector store stays up-to-date with files you drop into Drive.

4. Chat / Query Path

(Triggered by user chat via webhook)

• When a chat message arrives via When Chat Message Received, it gets passed into the AI Agent node.
• Before generation, the AI Agent calls the Pinecone Vector Store1 set in “retrieve-as-tool” mode, which runs a vector-based retrieval using the user query embedding. The relevant text chunks are pulled as tools/context.
• The OpenAI Chat Model node is linked as the language model for the agent.
• Simple Memory node provides conversational memory (keeping history across messages).
• The agent combines retrieved context + memory + user input and instructs the model to produce a response.

5. Connections / Flow Logic

• The Embeddings OpenAI node’s output is wired into Pinecone Vector Store (insert) and also into Pinecone Vector Store1 (so the same embeddings can be used for retrieval).
• The AI Agent has tool access to Pinecone retrieval and memory.
• The Download File node triggers the insert path.
• The When chat message triggers the agent path.

Similar Workflows / Inspirations & Comparisons

To help understand how your workflow fits into what’s already out there, here are a few analogues:

• n8n Blog: “Build a custom knowledge RAG chatbot” — they show a workflow that ingests documents from external sources, indexes them in Pinecone, and responds to queries via n8n + LLM. (n8n Blog)
• Index Documents from Google Drive to Pinecone — this is nearly identical for the ingestion part: trigger on Drive, split, embed, upload. (n8n)
• Build & Query RAG System with Google Drive, OpenAI, Pinecone — shows the full RAG + chat logic, same pattern. (n8n)
• Chat with GitHub API Documentation (RAG) — demonstrates converting API spec into chunks, embedding, retrieving, and chatting. (n8n)
• Community tutorials & forums talk about using the AI Agent node with tools like Pinecone, and how the RAG part is often built as a sub-workflow feeding an agent. (n8n Community)

What sets your workflow apart is your explicit combination: Google Drive → automatic ingestion → chat agent with tool integration + memory. Many templates show either ingestion or chat, but fewer show them combined cleanly with n8n’s AI Agent.

Suggested Published Description (you can paste/adjust)

> RAG AI Agent for Google Drive Documents (n8n workflow) > > This workflow turns a Google Drive folder into a live, queryable knowledge base. Drop PDF, docx, or text files into the folder → new documents are automatically indexed into a Pinecone vector store using OpenAI embeddings → you can ask questions via a webhook chat interface and the AI agent will retrieve relevant text, combine it with memory, and answer in context. > > Credentials needed > > * Google Drive OAuth2 (see: https://docs.n8n.io/integrations/builtin/credentials/google/oauth-generic/) > * Pinecone (see: https://docs.n8n.io/integrations/builtin/credentials/pinecone/) > * OpenAI (see: https://docs.n8n.io/integrations/builtin/credentials/openai/) > > How it works > > 1. Drive trigger picks up new files > 2. Download, split, embed, insert into Pinecone > 3. Chat webhook triggers AI Agent > 4. Agent retrieves relevant chunks + memory > 5. Agent uses OpenAI model to craft answer > > This is built on the core RAG pattern (ingest → retrieve → generate) and enhanced by n8n’s AI Agent node for clean tool integration. > > Inspiration & context > This approach follows best practices from existing n8n RAG tutorials and templates, such as the “Index Documents from Google Drive to Pinecone” ingestion workflow and “Build & Query RAG System” templates. (n8n) > > You’re free to swap out the data source (e.g. Dropbox, S3) or vector DB (e.g. Qdrant) as long as you adjust the relevant nodes.

If you like, I can generate a polished Markdown README for you (with badges, diagrams, instructions) ready for GitHub/n8n community publishing. Do you want me to build that?

🔗 Nodes Used

Google Drive, Google Drive Trigger, AI Agent, Embeddings OpenAI, OpenAI Chat Model, Simple Memory

📥 Import

Download workflow.json and import into n8n: Workflow menu → Import from File

📖 Importing guide · 🔑 Credential setup