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

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. (Pinecone1) 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. (n8n2) --- How to get the required credentials | Service | Purpose in Workflow | Setup Link | What you need / steps | | ------------------------- | ------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------- | | Google Drive (OAuth2) | Trigger new file events & download the file | https://docs.n8n.io/integrations/builtin/credentials/google/oauth-generic/ | Create a Google Cloud OAuth app, grant it Drive scopes, get client ID & secret, configure redirect URI, paste into n8n credentials. | | Pinecone | Vector database for embeddings | https://docs.n8n.io/integrations/builtin/credentials/pinecone/ | Sign up at Pinecone, in dashboard create an index, get API key + environment, paste into n8n credential. | | OpenAI | Embeddings + chat model | https://docs.n8n.io/integrations/builtin/credentials/openai/ | Log in to OpenAI, generate a secret API key, paste into n8n 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 ou