The digital virtual agent focuses on text-based interactions, with configuration and usage centered on the dialogue logic itself.
1. Creation & Configuration
Unlike voice agents, digital agents do not require configuration for voice (TTS) or speech recognition (ASR).
Path: AI Center → Digital Virtual Agent → Create New
-
Basic Information:
-
Name: Assign a name to your digital agent, e.g., “Pre-sales Consulting Bot” or “WhatsApp After-sales Agent.”
-
Conversation Context: Define the agent’s role, background, and overall objectives. This is identical to the setup for voice agents and is crucial for defining the AI’s core capabilities.
-
AI Engine / Model Version: Select the large language model that powers the AI’s reasoning, same as in voice agent configuration.
-
Description: Add notes on the agent’s intended purpose.
-
2. Workflow Design
The workflow design of a Digital Virtual Agent is the module responsible for execution (LLM + Workflow).
2.1 Trigger
Operation Path:
After creating a digital agent, click to enter the configuration page. By default, the Workflow Diagram tab is displayed.
Digital agents are typically used to handle inbound messages.
After the start node of the workflow, a Trigger node can be connected.
-
One or more keywords can be configured.
-
The workflow is activated only when the user’s first message contains the specified keywords.
-
If no keywords are configured, any new conversation sent to this channel will be handled by the digital agent.
-
Matching modes:
-
Fuzzy match
-
Exact match
-
2.2 Conditional Routing
Multiple matching rules can be configured.
Different branches can be executed based on the evaluation of variable values using:
-
Fuzzy matching
-
Exact matching
-
Regular expression (Regex) matching
This allows the workflow to identify different conditions and route the conversation accordingly.
2.3 Execution Time Window
Used to define execution rules based on a specific time zone, with configuration by:
-
Different days of the week
-
Specific time ranges
Operations will only be executed within the defined time windows.
2.4 Percentage Allocation
Distributes incoming traffic to different branches according to predefined ratios.
Commonly used for A/B testing or traffic allocation scenarios.
2.5 Jump
By identifying user intent, the workflow can skip unnecessary intermediate branches and jump directly to the target node, making interactions more efficient and seamless.
2.6 Add / Remove Tags
Allows adding or removing tags during the workflow execution.
2.7 Transfer
Supports selecting a transfer target:
-
Transfer to a specific agent group
Pre-transfer prompt:
-
Two options available:
-
Text-to-Speech (TTS)
-
Script (predefined content)
-
2.8 End Conversation
Ends the current conversation.
2.9 Send Message
Supports sending messages using text scripts.
2.10 Wait for Reply
Pauses the workflow and waits for the user’s response.
2.11 Invoke API
Fill in the corresponding values based on your HTTP request parameters.
-
The response data can be retrieved after the request is completed.
-
JSONPath syntax is supported for extracting response fields.
2.12 Modify Variable
Supports updating the value of one variable to another value.
2.13 Delayed
-
Delay Name: Allows entering a name for the delayed execution.
-
Delay Duration: Supports configuring delay time in:
-
Days
-
Hours
-
Minutes
-
Seconds
-
2.14 AI Agent Conversation
-
Name: Set a descriptive name for the node.
-
Prompt: Define instructions for the LLM. You can guide it as if you were training an employee.
-
Optimization: Provides prompt optimization suggestions for reference only. Prompts should be carefully tested and adjusted to meet your requirements.
-
Fixed Opening Message: Sets a fixed opening statement when this node starts.
-
Force Send: When enabled, ensures the fixed opening message is always sent. Otherwise, the node may be skipped if the current conversation already meets branching conditions.
-
Branch Name & Description:
Identifies conditions based on the full conversation. When the description is matched, the system exits the current multi-turn dialogue and proceeds to the next node of the corresponding branch.
Ensure branch descriptions are clear and independent. -
Path Tag:
When enabled, exit branches are tagged, and the branch name and description are saved as conversation tags, which can be analyzed in conversation logs and reports. -
Decision Description:
Assists the agent in judging specific situations.
Example: If the outstanding amount is 2000 and the customer pays 1000 → partial payment; if 2000 → full payment. -
Maximum Turns:
Sets the maximum number of dialogue turns. If consecutive timeouts occur or the limit is exceeded, the conversation exits early. -
Global Jump:
When enabled, allows any other intelligent conversation node to jump to this node.
Used for handling common off-SOP questions (e.g., “Who are you?”). -
(Global) Branch Name & Description:
Can be matched at any point in the conversation, enabling more flexible handling of “other” topics. -
When Leaving the Current Conversation:
-
Jump to another node: Jump directly to a specified node when exiting the global node.
-
Return to previous node: Resume the previous conversation flow after exiting the global node.
-
2.15 AI Collection
Information Collection:
-
Select variables to be collected (multiple selection supported)
-
Configure whether each variable is required
Advanced Settings:
-
Maximum Turns:
If the user does not respond within the maximum number of turns, the node ends and proceeds to the next step. -
Randomness (Temperature):
Controls the response creativity of the AI model:-
Lower temperature → More stable, conservative, and consistent responses
-
Higher temperature → More diverse, creative, and unpredictable responses
-
2.16 AI Agent Summary
-
Name: Enter the name of the node
-
Expected Intent Tags: Select intent tags to assign to the user
-
Prompt: Enter the summarization prompt
-
Description: Provide a description for this node
3. Experience Testing
After completing the workflow design, you need to thoroughly test the digital agent through text-based testing to verify that the logic meets expectations.
Operation Path:
On the digital agent editing page, click the Test button in the top-right corner.
-
The test interface is a chat window where you can simulate user interactions by entering text.
-
Conversation Flow: Verify whether the agent’s responses are accurate and smooth.
-
Debug Mode:
Click the debug icon to view real-time node transitions and variable value changes, which is highly useful for troubleshooting logic issues.
4. Channel Deployment & Management
After configuration and testing, the digital agent must be deployed to actual business channels before it can start serving users.
Example: WhatsApp Channel
Operation Path:
Management → Digital Channels → Channel Settings → WhatsApp
Enable Agent Hosting
-
In the channel list, locate the account where you want to deploy the agent.
-
Enable the Agent Hosting toggle.
-
Select the tested Digital Virtual Agent from the dropdown list.
Note: Once enabled, the digital agent will take over the reception flow for this account, replacing the original auto-reply logic.
Start Service
-
After saving the settings, all new messages sent to this WhatsApp account will be handled by the digital agent with priority (subject to trigger conditions).
-
If the agent cannot resolve the issue or a handoff condition is triggered, the conversation will be routed to human agents according to predefined rules.
5. View Conversation Records
Similar to voice agents, all conversations handled by digital agents can be reviewed and analyzed.
Operation Path:
Reports → AI Conversations
In the AI Conversations section, you can:
-
Filter and view conversations handled by AI
-
Review complete conversation histories
-
Analyze AI intent tags, customer information, and other details
These records support ongoing analysis and optimization.