n8n AI Support Assistant with GPT-4 Mini and Pinecone

This n8n workflow demonstrates how to create an AI-powered support assistant that can answer questions based on a custom knowledge base stored in Pinecone, using an OpenAI Chat Model (like GPT-4 Mini) and Langchain agents. The workflow is triggered by an incoming webhook, processes a user query, retrieves relevant information from Pinecone, generates a response using the AI model, and sends the answer back via the webhook.

What it does

This workflow automates the process of querying a knowledge base and generating AI-driven responses:

1. Receives a User Query: It listens for incoming HTTP requests (webhooks) containing a user's question.
2. Prepares the Query: A "Set" node ensures the incoming data is structured correctly for the AI agent.
3. Initializes AI Agent: An "AI Agent" node (Langchain Agent) is configured to use a chat model, memory, and a Pinecone vector store as a tool.
   • OpenAI Chat Model: Utilizes an OpenAI chat model (e.g., GPT-4 Mini) for conversational AI.
   • Simple Memory: Maintains a short-term conversational memory to provide context for follow-up questions.
   • Pinecone Vector Store: Acts as a tool for the AI agent to search and retrieve relevant documents from a custom knowledge base.
     • OpenAI Embeddings: Used by the Pinecone Vector Store to convert text into numerical vectors for similarity searches.
     • Recursive Character Text Splitter: Used to break down large documents into smaller, manageable chunks before embedding and storing them in Pinecone.
     • Default Data Loader: A placeholder for loading documents into the vector store (this workflow focuses on the query part, not the initial data loading).
4. Processes the Query: The AI Agent uses its configured tools and model to understand the query, search Pinecone for answers, and formulate a coherent response.
5. Responds to Webhook: The generated AI response is sent back as the payload of the initial webhook request.

Prerequisites/Requirements

To use this workflow, you will need:

• n8n Instance: A running instance of n8n.
• OpenAI API Key: For the OpenAI Chat Model and Embeddings.
• Pinecone Account and API Key: To host your vector database. You'll also need a Pinecone Environment and Index Name.
• Data for Pinecone: Prior to running this workflow for queries, you need to have your knowledge base documents embedded and stored in a Pinecone index. This workflow assumes the Pinecone index is already populated.

Setup/Usage

1. Import the Workflow: Download the JSON provided and import it into your n8n instance.
2. Configure Credentials:
   • OpenAI: Set up your OpenAI API credentials in n8n.
   • Pinecone: Set up your Pinecone API credentials in n8n, including your API Key, Environment, and the name of your Pinecone index.
3. Activate the Webhook: Enable the "Webhook" trigger node. This will generate a unique URL.
4. Populate Pinecone (if not already done): This workflow focuses on querying. You would typically have a separate workflow or script to load your documents, split them, generate embeddings using OpenAI, and upsert them into your Pinecone index.
5. Test the Workflow:
   • Send an HTTP POST request to the Webhook URL.
   • The request body should be a JSON object containing your query, for example:

     {
         "query": "What are the benefits of using n8n for automation?"
     }
     

   • The workflow will process the query and return the AI-generated answer in the webhook response.
6. Customize: Adjust the AI Agent's prompt, temperature, and other parameters in the "AI Agent" node to fine-tune its behavior. You can also modify the "Set" node to handle different incoming data structures.