Chat with Internal Documents using Ollama, Supabase Vector DB, and Google Drive

This n8n workflow enables you to interactively chat with your internal documents stored in Google Drive. It leverages Ollama for language model capabilities, Supabase as a vector database for efficient document search, and n8n's Langchain integration to orchestrate the entire process.

The workflow simplifies the process of ingesting documents from Google Drive into a vector database and then using an AI agent to answer questions based on the content of these documents.

What it does

This workflow consists of two main parts: document ingestion and chat interaction.

Document Ingestion (triggered by Google Drive changes):

1. Monitors Google Drive: The Google Drive Trigger node listens for new or updated files in a specified Google Drive folder.
2. Extracts File Content: The Extract from File node reads the content of the newly added or modified document.
3. Splits Text: The Character Text Splitter node breaks down the document content into smaller, manageable chunks (for better vector embedding and search).
4. Generates Embeddings: The Embeddings Ollama node creates vector embeddings for each text chunk using the Ollama language model.
5. Stores in Supabase: The Supabase Vector Store node stores these embeddings and their associated text chunks in your Supabase vector database, making them searchable.

Chat Interaction (triggered by a chat message):

1. Listens for Chat Messages: The When chat message received node acts as the entry point for user queries.
2. Initializes AI Agent: The AI Agent node sets up a conversational AI agent.
3. Configures Chat Model: The Ollama Chat Model node provides the conversational AI capabilities using Ollama.
4. Configures Memory: The Postgres Chat Memory node maintains the conversation history, allowing the AI agent to remember previous interactions.
5. Provides Document Search Tool: The Vector Store Question Answer Tool node enables the AI agent to search the Supabase vector database for relevant document chunks based on the user's query.
6. Responds to User: The Respond to Webhook node sends the AI agent's answer back to the user.

Prerequisites/Requirements

To use this workflow, you will need:

• n8n Instance: A running n8n instance.
• Ollama Instance: A self-hosted or accessible Ollama instance with the desired language model (e.g., llama2, mistral) installed.
• Supabase Account: A Supabase project with a configured vector database. You will need your Supabase URL and API Key.
• Google Drive Account: A Google account with access to Google Drive.
• n8n Credentials:
  • Google Drive OAuth2 credentials.
  • Supabase API Key credentials.
  • Ollama API credentials (if required by your Ollama setup).
  • Postgres credentials for chat memory (if using a separate Postgres instance, otherwise Supabase's built-in Postgres can be used).

Setup/Usage

1. Import the workflow: Download the provided JSON and import it into your n8n instance.
2. Configure Credentials:
   • Google Drive: Set up your Google Drive OAuth2 credential in n8n.
   • Supabase: Set up your Supabase API Key credential in n8n, providing your Supabase URL and API Key.
   • Ollama: Configure your Ollama credentials in n8n. This might involve setting the base URL for your Ollama instance.
   • Postgres (for chat memory): If you're using a separate Postgres for chat memory, configure its credentials. If you intend to use Supabase's built-in Postgres, ensure the connection details are correctly configured within the Postgres Chat Memory node.
3. Configure Google Drive Trigger:
   • Select your Google Drive credential.
   • Specify the Google Drive folder you want to monitor for documents.
4. Configure Supabase Vector Store:
   • Select your Supabase credential.
   • Ensure the table name and embedding column match your Supabase vector database setup.
5. Configure Ollama Nodes:
   • In the Embeddings Ollama node, select your Ollama credential and specify the embedding model you want to use.
   • In the Ollama Chat Model node, select your Ollama credential and specify the chat model you want to use.
6. Configure Postgres Chat Memory:
   • Select your Postgres credential (or Supabase Postgres credential).
   • Ensure the table name for chat history is correctly configured.
7. Activate the workflow: Once all credentials and nodes are configured, activate the workflow.

Now, any new or updated documents in your specified Google Drive folder will be automatically processed and added to your Supabase vector database. You can then interact with the AI agent by sending chat messages to the webhook URL provided by the When chat message received node (or through a custom chat interface connected to it).