How to Set Up the Customer Agent

The Customer Agent is a feature that lets AI respond automatically to your customers on LINE. You add it as a Workflow action and configure the AI's role, the Knowledge Base it references, and what happens after a conversation ends. Once you tell it your response policy, the AI answers based on the nature of each question and hands off to a human only when needed.

Adding the Customer Agent to a Workflow

The Customer Agent runs as a Workflow action, so you combine it with a trigger such as receiving a message or a friend being added.

1. Open the Workflow editor
2. Add an action and select "Customer Agent"
3. The settings panel appears on the right

When you configure the AI's response policy and Knowledge Base in the settings panel, the Customer Agent starts a conversation with the customer the moment the trigger fires.

Defining the AI's role with the System Prompt

You define the AI's basic behavior and personality in the "System Prompt", and simply writing the role and response policy there is enough for the AI to answer accordingly.

A customer service template is included from the start, so we recommend using it as is at first and rewriting it in your own words as needed. If you want to revert, use "Reset to default" to return to the initial state.

The more specific your System Prompt, the more consistent the quality of the answers.

Setting the display name, avatar, and greeting message

You shape the AI's appearance and first message as seen by your customers, and there are three items to configure.

Preparing a message for when the AI cannot answer

"Message when AI cannot answer" is the text sent to the customer when the AI could not produce an answer. If you leave it blank, a default English message is sent instead, so always fill it in when you run the agent in another language.

This message goes out in three cases: when an error occurs, when the response doesn't come back in time, and when the conversation reaches its exchange limit. Writing it so it also says what to do next — something like "Sorry, we can't answer automatically right now. Please try again in a little while." — keeps the customer from getting stuck.

Saving questions and answers to HubSpot

You can keep what a customer asked the AI, and what the AI answered, on HubSpot contact properties. These fields appear in the settings panel only when the HubSpot integration is set up.

Only the most recent one is saved in each case, and the previous content is overwritten. Use it when you want "what they asked" to drive a HubSpot segment (list) or workflow.

Referencing a Knowledge Base

You choose the information source the AI uses for its answers from the "Knowledge Base", and when you select a document you registered in advance, the AI answers based on its contents.

You can select multiple documents at once. The more frequently asked questions and product information you register, the more accurately the AI can answer.

The Knowledge Base must be uploaded in advance. For how to register it, see "How to Manage the Knowledge Base".

Enabling the available tools

In addition to the Knowledge Base, you can enable "Enabled Tools" that let the AI look up information on the spot, and the AI automatically chooses the appropriate one from the enabled tools based on each question.

Tools for external services appear only when that integration is complete.

Designing the post-conversation branches (True/False)

The Customer Agent action has two branches, True and False, and where it proceeds changes depending on the customer's action or the AI's judgment, so placing actions after each branch lets you automate what happens once a conversation ends.

The end button and the human handoff button are shown to the customer as quick replies during the conversation.

Changing the button labels

You can change the wording of the buttons shown on LINE to clearer terms.

Setting the message shown on timeout

When one hour passes since the customer's last message, the conversation session ends automatically, and you can choose whether to send a message on timeout.

When you enable "Send a message after 1-hour timeout", you can send the customer any message you like when the session ends. Setting up a notice such as "The conversation has ended because time has elapsed" makes it easier for the customer to understand the situation.

Integrating with an external AI agent (A2A)

This is an advanced setting that connects an AI agent you developed in-house to the Customer Agent. "External AI Agents" is a feature that integrates with an external AI agent using the A2A protocol, and you don't need to configure it for normal use.

If you don't configure an external agent, the Customer Agent operates on its own.

Adding an external agent

You add external agents from the settings panel, and you can register multiple agents.

1. In the settings panel, click "Add External Agent"
2. Enter the following items

For how A2A integration works, see "Overview of A2A Integration".

Tips for effective use

The Customer Agent's answer quality and how smoothly handoffs go depend on how you build your settings, so getting the following three points in order before you start operating is effective.