Tunneling: live chat example on AIsuru

Tunneling: a live chat example

In this section we'll see how to implement a live chat using dynamic intents and tunneling. This example will let us explore both how tunneling works and the structure of dynamic intent calls in detail.

The live chat scenario

A live chat is a perfect example to illustrate tunneling because it requires:

• An ongoing conversation with an external service;

• Maintaining context between messages;

• Handling responses in real time.

Dynamic intent structure

To implement the live chat, we configure the dynamic intent with:

• Name: "LIVE_CHAT";

• Activation phrases: "I want to speak to an operator", "Live chat", etc.;

• Webhook: URL of the service that manages the live chat.

When the user triggers the intent, the webhook receives a request with this structure:

{
  "intentName": "LIVE_CHAT",
  "utterance": "I want to speak to an operator",
  "slotValues": {},
  "currentTag": "string",
  "currentTagAuthenticated": true,
  "contextVars": {},
  "memoriID": "string",
  "sessionID": "string",
  "culture": "en-US"
}

Starting the tunneling

The webhook responds by activating tunneling to take control of the conversation:

{
  "emission": "I'm connecting you with an operator. Please wait a moment...",
  "conclusive": true,
  "tunneling": true
  }
}

From this point:

1. Every user message is sent directly to the webhook;

2. The Agent no longer processes messages — it just forwards them to the chat service;

3. The operator's replies arrive through the webhook (which passes them to the Agent via the emission).

Managing the conversation

During the chat, each user message generates a request to the webhook:

{
  "intentName": "LIVE_CHAT",
  "utterance": "I have a problem with my order",
  "sessionID": "string",
  "culture": "en-US"
}

The webhook responds while keeping the tunneling active:

{
  "emission": "Operator: Of course, please tell me your order number",
  "tunneling": true,
  "conclusive": true
}

Closing the chat

When the operator closes the chat, the webhook sends a response that ends the tunneling:

{
  "emission": "The chat has been closed. Thank you for using our support service!",
  "tunneling": false,
  "conclusive": true
}

The Agent resumes normal control of the conversation.

The tunneling flag determines who controls the conversation: if true, the webhook is in control; if false, the Agent takes back control.

The live chat example shows how tunneling lets you create complex interactions while keeping a clean, manageable architecture. The same principles apply to many other scenarios that require an ongoing conversation with an external system.