Automate Telegram Support Handover from AI to Humans with GPT-4 and Email Alerts

This n8n workflow automates the process of handling support requests received via Telegram, leveraging an AI agent for initial responses and seamlessly handing over to a human agent with an email alert when necessary. This ensures efficient customer service by offloading routine queries to AI while providing a clear escalation path for complex issues.

What it does

1. Listens for Telegram Messages: The workflow is triggered by incoming messages to a configured Telegram bot.
2. Processes with AI Agent: The incoming message is fed into an AI Agent (powered by a Chat Language Model like Azure OpenAI or OpenRouter) with a simple memory to maintain context.
3. Determines Handover Need: The AI agent's response is analyzed to determine if a human handover is required. This is typically based on specific keywords or a designated "handover" action from the AI.
4. Routes Based on Handover Decision:
   • If no handover is needed: The AI's response is sent back to the user via Telegram.
   • If a handover is needed:
     • An email alert is sent to a human support team, including the customer's query and the AI's interaction history.
     • A confirmation message is sent to the user on Telegram, informing them that a human agent will follow up.
5. Logs Interactions (Optional): A "Data table" node is included, which can be configured to log interactions for analysis or auditing purposes.

Prerequisites/Requirements

• n8n Instance: A running n8n instance.
• Telegram Bot: A Telegram bot token and a chat ID to listen for messages and send responses.
• AI Language Model:
  • Azure OpenAI Service: An Azure OpenAI deployment with an appropriate chat model (e.g., GPT-4) and API credentials.
  • OR
  • OpenRouter Account: An OpenRouter API key and a selected chat model.
• SMTP Server: Access to an SMTP server for sending email alerts to human support agents.

Setup/Usage

1. Import the Workflow: Import the provided JSON into your n8n instance.
2. Configure Telegram Trigger:
   • Select your Telegram Bot API credential.
   • Ensure the webhook is set up correctly to receive messages.
3. Configure AI Agent:
   • AI Agent (Node ID: 1119):
     • Select the appropriate Language Model (Azure OpenAI Chat Model or OpenRouter Chat Model).
     • Configure the System Message to define the AI's role and instructions, including how it should indicate a handover (e.g., by outputting a specific phrase like "HANDOVER_TO_HUMAN").
     • Adjust the Tools if the AI needs to interact with other services.
   • Simple Memory (Node ID: 1163): This node maintains conversation context. No specific configuration is usually needed beyond connecting it.
   • Azure OpenAI Chat Model (Node ID: 1253) / OpenRouter Chat Model (Node ID: 1281):
     • Select your respective API credentials.
     • Configure the model name and other parameters as needed.
4. Configure "If" Node (Node ID: 20):
   • Modify the condition to check the AI Agent's output for the specific phrase or indicator that signals a human handover. For example, {{ $json.text.includes("HANDOVER_TO_HUMAN") }}.
5. Configure "Send Email" Node (Node ID: 11):
   • Set up your SMTP credential.
   • Specify the recipient email address(es) for human support.
   • Customize the email subject and body to include relevant information from the Telegram message and AI interaction (e.g., {{ $json.chat.id }}, {{ $json.text }}, {{ $node["AI Agent"].json.text }}).
6. Configure "Telegram" Nodes (Node IDs: 49 and 50):
   • Ensure the Telegram credential is selected for sending messages.
   • Customize the messages sent back to the user for both AI responses and handover confirmations.
7. Activate the Workflow: Once configured, activate the workflow to start processing Telegram messages.

Note on AI Handover Logic: The If node (ID 20) is crucial for determining when to escalate. The AI Agent should be prompted to output a specific, easily identifiable string (e.g., "HANDOVER_TO_HUMAN") when it decides a human is needed. This string can then be checked by the If node to branch the workflow accordingly.