Skip to main content

Integrating any AI without coding

Introduction​

In this tutorial, I will explain how to build an AI agent in Live Helper Chat without coding, using any third-party AI service as the model provider.

Live Helper Chat provides the harness - the conversation, the triggers, the tools (actions) and the state - while the external AI service provides the reasoning. The bot interface is how you wire the two together.

How it works​

The workflow is simple:

  • Define AI REST API calls - these are the agent's connection to the model provider.
  • Define a bot that uses them - this is the agent itself, with its triggers, events and tools.
  • Upon starting a conversation, set a uuid which acts as the current conversation handler for the third-party AI - this is the agent's memory.

The REST API has only one request defined.

In the user parameters, we define an additional argument task_key, which is set directly in the bot and the request body.

In the output parsing, we have defined three possible combinations:

  • Standard message
  • Indication that the chat should be transferred to an operator
  • Indication to end the conversation

The bot itself needs only four triggers.

Chat started​

This trigger is always executed if a visitor sends a message or clicks a button replied to by the bot (the meta_msg message field).

The trigger structure defines which method to execute when a visitor sends a message and which trigger to execute based on the bot's response. The response types are defined in the REST API itself.

Bot message send​

This is a standard message received from the bot. Additionally, we set an additional chat variable based on the response, which we later reuse in the REST API call itself. This is required for the AI to know what the current conversation is about.

Transfer to Operator​

This simply transfers the chat to an operator.

Default response​

In this case, the default response acts as a conversation-ending trigger and displays a bot message.