Skip to main content

User guide

This guide explains how to use agent building and visual flow skill orchestration in Tongyi Xiaomi Conversational Agent.

1. Activate the product

image
  • Go to the Tongyi Xiaomi Conversational Agent console and click Enable for Free in the upper-right corner.
image
  • On the activation page for Tongyi Xiaomi Conversational Agent, activate the service. After activation, you are billed based on your actual call volume. Bills are generated daily and deducted from your Alibaba Cloud account balance.
image

2.Agent building

2.1 Create an application

Click Create Application, then enter an application name and description to create a basic agent.
  • Application Name: Enter a name for the agent based on your business requirements.
  • Application Description: Add a description for the agent based on its intended use case.
image
 After you create the application, the console is displayed as follows:
image

2.2 Configuration

1. Persona

  • Definition: A persona lets you shape your agent's behavior with a prompt. You can define its character, role, available skills, workflow, and other constraints.
    image
  • Select Template: In the Select Template window, you can insert a general template and customize the persona by filling in the required details.
    image
  • Optimize: After editing the persona, click Optimize. The model will then refine your persona description based on your input.
    image

2. Agent greeting

  • Definition: The agent sends this opening message when a user starts a conversation.
    image

3. Knowledge base (flow skills and FAQ)

4. More settings

  • Model: The service uses the built-in Tongyi Xiaomi large model. You can modify its behavior by adjusting the temperature and the number of context turns.
  • General Knowledge:
  • Response Language: The language the agent uses for its replies.
    • Same as user's language: The agent replies in the same language as the user's query.
    • Chinese only: The agent always replies in Chinese, regardless of the query language.
    • English only: The agent always replies in English, regardless of the query language.
  • Security:
    • Security Interception: If the system detects sensitive content in the input or output, it automatically replies with a preset security response.
    • Preset Security Response: When a user's query or the model's generated response contains sensitive content, the agent automatically sends this response. The default response is: "I cannot answer this question. Please try asking something else."
  • Model Generation Failure: The agent's fallback response for system errors or service timeouts.
image

2.3 Call volume

The call volume page shows your token consumption.
image

2.4 API

The agent's debugging information syncs to the API. You can copy the provided code to make calls.
image

3.Visual flow orchestration

3.1 Create a flow

  1. Click Knowledge Base Management > Flow or + Bind Flow to open the flow list window.
    image
    image
  2. Click New Flow to open the creation dialog box.
    image
  3. Enter a Name and Description for the flow, then click Confirm to create a blank flow.
    image
  4. Find the new flow and click Modify to open the flow orchestration page.
    image
    image
  5. Create a starting branch to enter the flow canvas. After the start node, add a branch and configure the conditions that trigger it. For detailed steps, watch the video below:
    image.gif
  6. Select and arrange nodes according to your conversation logic to build the dialogue flow.
    image.png
After you finish orchestrating the flow, click Test. The system automatically checks the flow for completeness. If the system finds any errors, follow the prompts to optimize the flow. For a detailed demonstration, watch the video below:
image.gif

3.2 Flow debugging

  1. Before debugging, configure the environment.
  • Service Simulation: An API plugin in the flow returns a mock value directly. This is useful when the actual API is not yet ready.
  • Passthrough Parameters: These are external parameters sent to the agent along with the user's message. For example, in a phone call scenario, you can pass the user's phone number as a passthrough parameter for use in subsequent API plugin calls.
image.png
  1. Conversation Debugging Start a conversation directly with the agent. After it replies, you can click Generation Complete to view the full output. You can also inspect the specific parameters that the agent gathered.
    image.gif

4. Configure FAQ

4.1 Create an FAQ base

  1. Click Knowledge Base Management > FAQ or + Bind FAQ to open the FAQ list window.
    image
    image
  2. Click Create FAQ Base and enter a name for it in the creation window.
    image
  3. Find the new FAQ base and click Edit to add FAQ entries.
    image
    image
  4. Click Add FAQ or Import FAQ to add questions. Fill in the required fields, such as Question, Answer Type, Answer, Effective Time, and Similar Questions, then click Submit to save the entry.
    image
    image
     After successfully adding an entry:
    image

4.2 Bind and test FAQ

  1. Click Bind to attach the knowledge base to the agent.
    image
After a successful binding:
image
  1. Test the result in the conversation test window on the left.
    image
Agent Development
Agent Context
Agent Governance & Evolution
Store
Application Marketplace
Use Cases
Support