n8n Supabase & Ollama Workflow Automation

1. Introduction

This document describes the end-to-end workflow you built in n8n for:

1. Downloading user documents from Google Drive.
   
2. Converting documents into vector embeddings using the Ollama embedding model.
   
3. Splitting text into manageable chunks.
   
4. Storing embeddings in Supabase Vector Store.
   
5. Accepting chat queries and answering via a Question-and-Answer (Q&A) chain.
   

Your LLM of choice (Ollama) powers both embeddings and response generation.

---

2. Architecture Overview

flowchart LR

    subgraph Upload Pipeline

      A[“Click Test workflow”] –> B[Google Drive: Download file]

      B –> C[Supabase Vector Store: EmbedDocument]

      C –> D[Default Data Loader]

      D –> E[Recursive Character Text Splitter]

      E –> C

      C –> F[Embeddings Ollama]

    end

    subgraph Q&A Pipeline

      G[“Chat message received”] –> H[Question and Answer Chain]

      H –> I[Vector Store Retriever]

      I –> J[Supabase Vector Store]

      H –> K[Ollama Model]

      J –> L[Embeddings Ollama]

    end

• Upload Pipeline: Triggered manually via the “Test workflow” button. Downloads files, generates and stores embeddings.
  
• Q&A Pipeline: Triggered by new chat messages, retrieves relevant embedding chunks, and generates answers.
  

---

3. Prerequisites

• n8n: Version ≥ 1.x with Google Drive and Supabase nodes installed.
  
• Supabase project: Create a Vector Store table with id, content, metadata, and embedding columns.
  
• Ollama: Running locally or accessible via API (http://localhost:11434 by default).
  
• Environment variables:
  
  • SUPABASE_URL
    
  • SUPABASE_SERVICE_KEY
    
  • OLLAMA_EMBED_URL
    
  • OLLAMA_MODEL_EMBED (e.g. llama2-7b)
    
  • OLLAMA_MODEL_CHAT (e.g. llama2-7b-chat)
    

---

4. Upload Pipeline Details

4.1 Trigger Node: “Test workflow”

• Type: Manual trigger
  
• Description: Kicks off the upload sequence for one or more files.
  

4.2 Google Drive: Download File

• Credentials: Google OAuth2 credentials
  
• Operation: download file
  
• Parameters:
  
  • File ID: Set via expression or hard-coded
    
  • Binary Property: data
    

4.3 Supabase Vector Store: EmbedDocument

• Credentials: Supabase Service Role Key
  
• Operation: EmbedDocument
  
• Parameters:
  
  • Document: Binary data from Google Drive
    
  • Embeddings: Use the Ollama embedding model node
    

4.4 Default Data Loader & Text Splitter

• Default Data Loader: Converts raw document into plain text
  
• Recursive Character Text Splitter:
  
  • Chunk Size: 1000 characters
    
  • Overlap: 200 characters
    

4.5 Ollama Embeddings Node

• Endpoint: {{ $env.OLLAMA_EMBED_URL }}/embeddings
  
• Model: {{ $env.OLLAMA_MODEL_EMBED }}
  

---

5. Q&A Pipeline Details

5.1 Trigger Node: Chat Message Received

• Type: Webhook / Chat integration
  
• Description: Listens for incoming user chat messages
  

5.2 Vector Store Retriever

• Vector Store: Supabase
  
• Retrieval Params:
  
  • Top K: 5
    
  • Filter: Optional metadata filters
    

5.3 Question and Answer Chain

• Model Node (Ollama Chat):
  
  • Endpoint: {{ $env.OLLAMA_EMBED_URL }}/chat
    
  • Model: {{ $env.OLLAMA_MODEL_CHAT }}
    
• Chain Type: RetrievalQA (combine retrieved chunks + user query)
  

---

6. Environment & Configuration

Variable	Description
SUPABASE_URL	Your Supabase project URL
SUPABASE_SERVICE_KEY	Service role key for full database access
OLLAMA_EMBED_URL	Base URL for Ollama server
OLLAMA_MODEL_EMBED	Embedding model name (e.g. llama2-7b)
OLLAMA_MODEL_CHAT	Chat model name (e.g. llama2-7b-chat)

---

7. Testing & Deployment

1. Manual Test: Click “Test workflow” in n8n UI to verify upload pipeline.
   
2. Chat Simulation: Send a dummy query via your chat UI to confirm Q&A pipeline.