🚀 Creating an RAG AI Agent in n8n

This repository contains a detailed tutorial for building an RAG (Retrieval-Augmented Generation) AI Agent in n8n, capable of answering questions based on documents stored in a vector database.

📌 Objective

Create an AI agent that can answer questions based on stored documents using GPT-4o mini from OpenAI, PostgreSQL for memory, and Supabase Vector Store for vector storage.

🛠️ Technologies Used

• n8n - Automation platform
• OpenAI GPT-4o mini - AI model
• PostgreSQL - AI agent memory
• Supabase Vector Store - Vector storage
• Google Drive - Document upload

🔗 Integration of Tools

The AI agent workflow involves:

1. Connecting OpenAI's API to PostgreSQL and Supabase Vector Store
2. Configuring the AI agent's memory in PostgreSQL
3. Storing documents in Supabase Vector Store
4. Creating a Google Drive trigger to capture new documents
5. Processing files and generating embeddings with OpenAI
6. Using n8n to manage the data flow

The test documents contain fictional questions and answers for a virtual Customer Support assistant, generated by ChatGPT.

---

📖 Step-by-Step Guide

1️⃣ Creating the AI Agent in n8n

• Add a message trigger
• Use the node "Advanced AI / AI Agent"
• The AI agent will respond and decide when to consult the database
• Configure the "AI Agent"
  • Chat Model: Create OpenAI credentials
  • Memory: Choose PostgreSQL (using Supabase)

2️⃣ Setting Up Supabase

• Create a new project in Supabase (e.g., n8n AI Agent - Date - DEMO Z)
• Set a password for the database
• Ensure the project is active (Check Project Status)
• Go to Project Settings > Database > Connection Parameters

3️⃣ Configuring PostgreSQL in n8n

• Create a new PostgreSQL credential in n8n
• Configure the parameters:
  • Host, Port, User, Password (obtained from Supabase)

4️⃣ Setting Up the Vector Store Tool

• Add the vector storage tool
• Name and description
• Select Supabase Vector Store
• Connect to the Supabase account (Project Settings > API)
• Copy URL and secret key into n8n
• Create the storage table
• In Supabase SQL Editor, run the script from n8n documentation (Quickstart for setting up your vector store)
• Choose Table Name: documents
• Add option: match_documents

5️⃣ Uploading and Processing Documents

• Create a folder in Google Drive
• In n8n, add a trigger "Google Drive - On Changes on Specific Folder"
• Select event: When a document is created
• Upload documents to the Google Drive folder
• Run "Fetch Test Event" in n8n to verify document detection

6️⃣ Processing the Files

• Convert the document to a processable format
  • Google File Conversion: TXT to CSV
• Download the document
  • Google Drive - Download File by ID
• Transform the file
  • Data Transformation: Extract from text file
• Ensure no duplicates in Supabase
  • Delete previous records in Supabase Vector Store
  • Add the condition operation: delete / filters (string):

    metadata->>file_id=like.*{{$json.id}}*

7️⃣ Creating Embeddings in Supabase Vector Store

• Add OpenAI embeddings
• Define chunk size and overlap for document splitting
• Use the "Recursive Character Test" model
• Save settings and test agent queries

---

📌 Final Considerations

Your AI agent is now ready to answer questions based on stored documents! 🚀
It can retrieve information from the vector database and generate intelligent responses using the GPT-4o mini model.

If you need assistance or want to contribute, feel free to open an issue or submit a pull request!