RAG Agent – Supabase + Gemini + n8n Workflow
This project is a fully automated RAG (Retrieval-Augmented Generation) AI Agent built using n8n, Google Gemini, Supabase Vector Store, and PostgreSQL chat memory. It allows you to upload documents, generate embeddings, store them in Supabase, and run an AI chat agent that responds with contextual knowledge retrieved from stored embeddings.
🚀 Features Google Gemini Chat Model for high-quality AI responses RAG-based contextual answers using Supabase vector search Document ingestion via Google Drive Automatic embeddings generation using the Gemini Embedding model Persistent chat memory stored in PostgreSQL Automated chat response workflow triggered by incoming messages n8n-based visual and modular orchestration
🧩 Architecture Overview This workflow integrates several powerful components to create an intelligent RAG agent:
-
Chat Trigger Listens for incoming chat messages (chatTrigger node) Sends the user query to the AI Agent
-
AI Agent (LangChain Agent) Receives and processes user queries through: Google Gemini Chat Model Supabase Vector Store (as a retrieval tool) PostgreSQL chat memory Internal reasoning & tool execution pipeline
-
Google Gemini Chat Model Generates natural language responses Processes context from the vector store and memory
-
Supabase Vector Store Stores: Document embeddings Retrieved chunks for RAG responses
Used in two modes: Insert Mode — Stores new document embeddings Retrieve Mode (Tool) — Searches for relevant context to answer queries
-
Default Document Loader Converts binary documents (PDF, DOCX, TXT, etc.) into text Prepares documents for embedding generation
-
Google Drive Integration Downloads source files directly from your Google Drive Sends them to the loader for processing
-
Gemini Embedding Model Generates embeddings for documents Embeddings are stored inside Supabase for vector search
-
PostgreSQL Chat Memory Stores long-term conversational data: Previous user queries Chat context Past responses
Conversation history for personalized answers
📌 How It Works (Workflow Overview) Document Processing Manual Trigger → Google Drive Download → Document Loader → Gemini Embeddings → Supabase Vector Store (Insert)
Chat Processing Chat Trigger → AI Agent → Gemini Chat Model → Supabase Vector Search (RAG) → PostgreSQL Memory → Final Answer