Build an AI Agent in n8n: Chat with Files from Supabase Storage and Google Drive

Automatically vectorize documents from a Supabase private bucket or Google Drive folder, then query them through an AI agent with natural-language Q&A

Who

Developers, data analysts, or business users with basic n8n and database experience who need AI-powered Q&A over document repositories

You get

Files are auto-parsed, chunked, embedded, and stored as vectors; users chat via a chat interface or webhook for context-aware answers grounded in uploaded documents

You need

n8n · supabase · google-drive · llamaparse

Steps

1. Create Supabase bucket and upload files

   Create a private bucket in your Supabase project, modifying the default name as needed. Upload document files to this bucket using the provided upload options. This bucket serves as one of the primary file sources for the workflow.

2. Configure database tables and vector extension

   Create a files table (tracking file names and storage IDs) and a documents table (storing vectorized document chunks) in Supabase. Execute the necessary SQL queries to enable vector matching capabilities (e.g., pgvector extension and a match_documents query function).

3. Import workflow and replace credentials

   Import this workflow from the n8n template library (template ID 4086). Review all credential placeholders across nodes and replace them with your own: supabaseApi (Supabase project URL and service-role key), openAiApi (OpenAI API key), googleDriveOAuth2Api (Google Drive OAuth2 client credentials), and httpHeaderAuth (LlamaParse API key, formatted as Authorization: Bearer <api_key>). Update the <project_id> placeholder in HTTP Request nodes to your actual Supabase project ID.

   Setup guidance · Editorial suggestion。Setup assistant checklist; verify credentials and project ID replacement before execution

   Copy prompt

   Please check all sticky notes marked ### Replace credentials. and ensure the following credentials are correctly configured:
   1. supabaseApi - Supabase project URL and service-role key
   2. openAiApi - OpenAI API key
   3. googleDriveOAuth2Api - Google Drive OAuth2 client credentials
   4. httpHeaderAuth - LlamaParse API key (Header: Authorization: Bearer <api_key>)
   Also replace <project_id> in all HTTP Request nodes with your actual Supabase project ID.

4. Configure Google Drive monitoring trigger

   In the Scenario 3 branch of the workflow, configure the Google Drive Trigger nodes (File Created / File Updated) to watch a specific Google Drive folder. Replace the example folder ID in folderToWatch with your actual target folder ID. Set the polling interval (default: every minute). When new files are created or updated, the workflow automatically downloads and processes them.

5. Configure the AI agent chat interface

   In the Scenario 2 branch, configure the AI Agent node's system prompt and model parameters. The AI Agent accepts input via Chat Trigger (When chat message received) or Webhook. Select gpt-4o-mini or gpt-4o in the OpenAI Chat Model node. Configure the Simple Memory node to use a custom session key (from the Merge node's session_id) to maintain conversation context across turns.

   Execution prompt · Source text。AI Agent1 node systemMessage field; defines agent behavioral constraint

   Copy prompt

   You are a helpful assistant.
   
   IMPORTANT Always use knowledge base to respond users.

6. Configure vector store and memory persistence

   Set the Supabase Vector Store node to retrieve-as-tool mode, connecting to the documents table with tool name knowledge_base and description Gather facts from knowledge base to respond user.. Configure both Insert into Supabase Vectorstore nodes (for Scenario 1 and 3) to use the text-embedding-3-small model with chunkSize 500 and chunkOverlap 200. Simple Memory uses PostgreSQL to store user interactions and manage conversation states.

7. Test and fine-tune

   After configuration, use the manual trigger (When clicking 'Test workflow') to test the Scenario 1 file processing flow, verifying that files download, parse, vectorize, and insert correctly. Send test messages via Chat Trigger or Webhook to verify the AI Agent returns relevant answers grounded in stored documents. Adjust model parameters, chunk size, or the system prompt based on response quality.

You are assisting with reproducing an n8n workflow template: an AI file-Q&A agent powered by Supabase and Google Drive. Confirm prerequisites in this order: 1) Supabase project, private bucket, files/documents tables, and vector extension are ready; 2) Four credentials (supabaseApi, openAiApi, googleDriveOAuth2Api, httpHeaderAuth) are obtained and configured; 3) Workflow JSON is imported and all <project_id> and folder IDs are replaced with actual values; 4) LlamaParse API key is set in HTTP Header Auth. Once confirmed, execute each scenario step by step.