Configuring a communication agent involves defining the context for its tasks, such as its background, objectives, and requirements, and linking it to a private knowledge base. Media configuration involves setting the voice, volume, and event handling, and associating a custom voice and a hotword library. This topic explains how to configure a communication agent.
Prerequisites
You have created a knowledge base and uploaded files to it.
Create an agent
In the Intelligent Contact Center console, navigate to LLM-based Communication>Communication Agent, and on the Communication Agent Management tab, click Create Agent. In the dialog box that appears, configure the basic information and click OK to create the agent.
After you create a communication agent, its status on the Communication Agent Management page changes to Constructing. Construction typically takes 10 to 30 minutes, but the actual duration may vary. After construction is complete, the status changes to Debugging.
Basic information includes Agent Name (required, up to 30 characters) and Agent Description (up to 1,000 characters).
Voice and call event configuration
On the Communication Agent Management page, click Voice and Call Event Configuration. In the dialog box that appears, configure the TTS Voice Configuration, Hotword Configuration, and Event Handling Configuration.
In the TTS Voice Configuration section, you can set and audition the voice. In the Hotword Configuration section, you can select or create a hotword library. In the Event Handling Configuration section, you can set parameters such as Silence Duration, Silence Trigger Model, Hang Up on Silence Configuration (hangs up after a specified number of silence events), the answering machine detection switch, Hang Up Immediately, and Maximum Call Duration. After configuring these settings, click Save.
When you select a voice in the TTS Voice Configuration section, the Voice dialog box appears. For Type, select custom voice or system preset voice. You can filter voices by name, suggested use case, and gender. The list displays the voice name, traits, suggested use case, specific scenario, gender, and supported languages. Click Audition to preview a voice. Select the one you want and click OK.
In the TTS Voice Configuration section, use the sliders to adjust the speech rate (range: -200 to 200) and volume (range: 0 to 100). In the mixing and background sound configuration area, choose whether to enable mixing and background sound. If you enable background sound, you can choose from Call Center, Office, or Typing Sound. Click Audition to preview the effect, and then click OK to save the configuration.
Parameters:
Parameter | Parameter | Description |
TTS Voice Configuration | Voice | Voice settings:
|
Content Audition: Customize the content for voice auditions. | ||
Hotword Configuration | Hotword Library | Configure a hotword library. If you do not have one, you must first create a hotword library. |
Event Handling Configuration | Silence Duration | If both parties are silent for the specified duration, a silence event is generated. Valid values: 3s to 15s. |
Silence Trigger Model | When the first silence event occurs, the system pushes the event to the model. | |
Hang Up on Silence Configuration | Specifies the number of consecutive silence events (from 1 to 5) that will trigger the system to end the call. | |
Answering Machine Detection | When enabled, this feature automatically detects voicemails or answering machines and ends the call to improve call efficiency. | |
Hang Up Immediately | Determines whether to end the call immediately upon detecting an answering machine or voicemail. | |
Maximum Call Duration | Sets the maximum duration for the call. The call automatically hangs up when the time limit is reached. |
Click Confirm to save the configuration.
Edit an agent
Prerequisite: The communication agent's status must be Debugging.
On the Communication Agent Management page, click Edit. Configure Basic Information, Agent Customization, Custom Exception Handling, Call Variable Configuration, Main Intent Configuration, and Output Tag Configuration.
Parameters:
Parameter
Parameter
Description
Example
Basic Information
Business Scenario
Select a scenario that matches your business.
Convert individual customer leads
Agent Role
Define the role of the agent.
Chang, a service advisor at an automotive company
Counterpart Role
Define the role of the other party in the conversation.
A customer interested in purchasing a car
Language Style
Select a language style appropriate for your business.
Friendly
Associated Knowledge Base
Select a relevant knowledge base, such as one containing information about car models.
Automotive Knowledge Base
Agent Customization
Task Type
Basic Task: Suitable for simpler scenarios. Configure the opening statement and task settings, and describe the task content and flow as clearly as possible.
Advanced Task: Suitable for scenarios that require breaking a task into multiple subtasks and managing each with fine-grained controls, such as separate opening statements or key scripts. This improves the determinism and stability of task execution.
Advanced Task
Opening Statement
Text: Define the text content for the opening statement.
Recording: Supports WAV, MP3, and M4A formats. The file size cannot exceed 2 MB, and the duration cannot exceed 60 seconds.
NoteIf you use a recording for the opening statement, we recommend using a cloned voice from the same speaker for a more natural sound.
Text: I'm the official advisor from ${name} Automotive. We're currently offering a 20% discount on all our products, which is a great deal. Would you be interested in learning more?
Task Configuration
Basic Task: Assign a task to the communication agent.
Advanced Task: Break the task into multiple subtasks and configure a task name and standard script for each.
Task Name
Invite the customer to visit a store for an in-person experience
Standard Script
If the customer agrees, ask for their city, district/county, or specific location in a step-by-step manner.
If the customer declines, try to retain them with a follow-up. Example script: "I understand. The best way to know if a product is right for you is to try it out, and it won't take long. I can arrange a visit to a nearby store, so you can stop by at your convenience."
Flow Description
Provide a detailed description of the execution flow and logic for the advanced task. Variables are supported. This option is available only for advanced tasks.
Use the opening statement to confirm if the customer wants to learn about the ${name} model.
Invite the customer to the store for a test drive.
Ask for address information.
End the conversation politely.
Custom Exception Handling
Custom Exception Handling
Configure exception-handling policies so that the communication agent replies as required when it encounters specific sensitive questions.
Enabled
Call Variable Configuration
Call Variable Configuration
The system automatically detects variables referenced in the opening statement, flow description, and task configuration. These variables can be manually edited, including their names and types.
The Call Variable Configuration page displays variable information in a table with four columns: Variable Key, Variable Name, Type/Possible Values, and Actions. For example, a variable key could be
name, with the variable name "Name" and possible values "XiaoKai Plus" and "XiaoHu Plus". You can click Edit to modify the variable configuration. The Re-identify function in the upper-right corner allows you to rescan the call configuration for variables.Main Intent Configuration
Main Intent Configuration
Configure the main intent of the call. The main intent is the core metric used to determine whether the call task was successful. The system provides a pre-generated template for the main intent, which you can modify to fit your business needs. Currently, the main intent only supports structured, single-choice outputs. You can edit the main intent name, description, and possible values.
The default main intent is named Call Effect Rating, with the description "Call Effect Rating: A, B, C, D (A is best)" and possible values of A, B, C, and D. You can click Edit to modify these settings.
Output Tag Configuration
Output Tag Configuration
Configure output tags (optional) to capture key information from the call process and outcome, supplementing the main intent. Clearly define the content and criteria for each tag. Output tags can be structured options or text.
This example shows six configured output tags: City (extracts the customer's city name, type: text), District/County (extracts the customer's administrative district or county name, type: text), Address (extracts specific landmarks, streets, or residential areas mentioned by the customer, type: text), Address-Longitude (extracts the longitude of the address, type: text), Car Purchase Budget (matches the customer's budget range with possible values like 100,000-200,000, 200,000-300,000, 300,000-400,000, 400,000-500,000; type: text), and Requirement Type (identifies the customer's purchase need with possible values like additional purchase, first purchase, trade-in, or ride-hailing; type: text).
Submit for review
After submission, the communication agent's status changes to Reviewing. The review process typically takes about one day.
Approved: The status changes to Ready to Go-live, and you can perform call tests.
Rejected: The status changes to Build Failed. Check the rejection reason and rebuild the agent.
After you finish editing the communication agent, click Submit for Review.
This option is available in the Actions column only for agents with a Debugging status.
In the dialog box that appears, click Confirm to submit.
Expert dial test
You can add up to 10 authorized numbers and initiate up to 5 dial test requests simultaneously.
After the communication agent is approved, click Expert Dial Test to test phone numbers.
On the expert dial test page, click Add Number.
A message at the top of the page indicates that you can add up to 10 authorized numbers and initiate up to 5 dial test requests at the same time. In the number list, you can Dial Test or Delete existing numbers. You can also use batch dial test to initiate multiple tests at once.
In the dialog box, authorize a mobile number for the dial test. You must enter a verification code to proceed. Click Confirm to add the number.
On the expert dial test page, select one or more numbers and click Dial Test. You can select up to five numbers for a batch dial test.
In the dialog box that appears, click Dial Test to start the test.
Launch agent
After the communication agent goes live, its status changes to Online.
Once the agent is online, you can create call tasks to initiate calls. For more information, see Call Task Management.
Prerequisite: The communication agent's status must be Ready to Go-live.
On the Communication Agent Management page, click Go-live.
In the dialog box that appears, click Confirm Go-live.
Take agent offline
You cannot take a communication agent offline if it is associated with existing call tasks.
On the Communication Agent Management page, click Go-offline.
In the dialog box that appears, click Confirm Go-offline.
Delete agent
This action cannot be undone. Proceed with caution.
You cannot delete a communication agent if its status is Constructing or Online.
On the Communication Agent Management page, click Delete.
In the dialog box that appears, click Confirm Deletion.
Next steps
After you configure a communication agent and it goes live, you can use it to configure call tasks. For more information, see Call Task Management.