Customer Support RAG Agent with GPT and Pinecone (Telegram Integration)

This n8n workflow automates a customer support RAG (Retrieval Augmented Generation) agent, allowing it to answer user queries from Telegram by leveraging a Pinecone vector store and an OpenAI GPT model. The agent is designed to provide intelligent and context-aware responses based on a knowledge base stored in Pinecone.

What it does

This workflow streamlines customer support interactions by:

1. Listening for Telegram Messages: It acts as a Telegram bot, waiting for incoming messages from users.
2. Processing User Queries with an AI Agent: When a message is received, it triggers an AI Agent (powered by LangChain) to process the user's query.
3. Retrieving Context from Pinecone: The AI Agent utilizes a Pinecone Vector Store to retrieve relevant information from a predefined knowledge base based on the user's query. This is augmented by OpenAI Embeddings to understand the semantic meaning of the query and the stored data.
4. Generating Responses with OpenAI Chat Model: The retrieved context, along with the user's query, is then fed into an OpenAI Chat Model (GPT) to generate a comprehensive and accurate response.
5. Maintaining Conversation History: A simple memory buffer is used to keep track of the conversation, allowing the AI agent to provide more coherent and contextually relevant responses over time.
6. Sending Responses via Telegram: The generated response from the AI agent is sent back to the user in the Telegram chat.

Prerequisites/Requirements

To use this workflow, you will need:

• n8n Instance: A running instance of n8n.
• Telegram Bot Token: A Telegram bot token obtained from BotFather.
• OpenAI API Key: An API key for OpenAI to access their embedding and chat models (e.g., GPT-3.5 or GPT-4).
• Pinecone Account and API Key: An account with Pinecone and its corresponding API key and environment details.
• Pinecone Index: A pre-populated Pinecone index containing your customer support knowledge base, with embeddings generated using the same OpenAI embeddings model configured in the workflow.

Setup/Usage

1. Import the Workflow: Download the provided JSON and import it into your n8n instance.
2. Configure Telegram Trigger:
   • Select your Telegram Bot credential.
   • Ensure the "Updates" are set to listen for messages.
3. Configure Telegram Node:
   • Select the same Telegram Bot credential.
   • The "Chat ID" should be set dynamically to {{ $json.chat.id }} to reply to the originating chat.
   • The "Text" field should be set dynamically to {{ $('AI Agent').item.json.response }} to send the AI agent's response.
4. Configure OpenAI Embeddings:
   • Select your OpenAI API credential.
   • Choose the appropriate embedding model (e.g., text-embedding-ada-002).
5. Configure Pinecone Vector Store:
   • Select your Pinecone API credential.
   • Provide your Pinecone environment and the name of your pre-existing Pinecone index.
6. Configure OpenAI Chat Model:
   • Select your OpenAI API credential.
   • Choose your desired chat model (e.g., gpt-3.5-turbo or gpt-4).
7. Configure Simple Memory:
   • This node manages the conversation history. You can adjust the k parameter to control how many previous messages are remembered.
8. Configure AI Agent:
   • Ensure the "Tools" are connected to the "Pinecone Vector Store" node.
   • Ensure the "Language Model" is connected to the "OpenAI Chat Model" node.
   • Ensure the "Memory" is connected to the "Simple Memory" node.
   • The "Input" should be set dynamically to {{ $json.text }} to feed the Telegram message into the agent.
9. Activate the Workflow: Once all credentials and configurations are set, activate the workflow.

Your Telegram bot will now be ready to answer customer support queries using your Pinecone knowledge base and OpenAI's intelligence!