Previewer

• Rapidly iterate on your GenerativeAgent’s design
• Test capabilities in real time before deploying to production
• Validate how your agent responds to different scenarios and environments

​

Using the Previewer

1

Select Previewer Environment

Choose the environment your GenerativeAgent uses for testing and previewing conversations.Select one of the following:

• Draft
• Sandbox
• Production

The Previewer Environment is set to Draft by default.

2

Select Channel Type

• Voice
• Messaging

Speak to GenerativeAgent in your browser to simulate voice calls. This is useful to assess how your GenerativeAgent responds in real time, including spoken utterances and actions.Conversation controls:

• Mute Self: Exclude your audio from the simulation.
• Mute GA: Mute GenerativeAgent speech while still showing the transcript.
• End Call: End the voice session.

3

Select Preview Mode

• Talk to GenerativeAgent
• Simulate a customer and conversation

Manually interact with your GenerativeAgent.

This is the only mode available for voice conversations.

You can choose:

• Test Scenario Type: Use data from a previously created Test Scenario to simulate function calls.
  • This allows you to iterate on tasks definitions or on Functions without concern of hitting actual APIs.
• External Endpoint: Use actual API connections to test with real data.
  Most preview testing uses test scenarios as it is faster to design iterations but the External Endpoint is helpful for final QA testing and pre-launch validation.

When using the External Endpoint, provide:

• User ID: Required for ASAPP APIs.
• Task Name: The specific task for the GenerativeAgent.
• Input Variables: Input variables data in JSON format.
  Input variables can be submitted as key-value pairs in JSON format.
  Input Variables Example

4

Enable Human-in-the-loop Agent (optional)

HILA is natively supported in the Previewer through a dedicated panel for simulating and validating scenarios where your GenerativeAgent requests assistance.

1. Enable HILA Support Go to GenerativeAgent Settings in the top right corner and toggle on Expert Assistance.

2. Simulate a Conversation Start a conversation where your GenerativeAgent is expected to request assistance.
3. Trigger the HILA Panel When assistance is requested, the HILA panel appears, allowing you to step in as an agent.
4. Respond as an Agent Use the panel to simulate multi-turn interactions and test agent guidance.
5. Expand the View (optional) Pop out the HILA panel into a separate browser tab for easier access and a larger workspace.

5

Customize Previewer Layout (optional)

• Full Screen: Click the expand icon in the top-right corner to enlarge the Previewer.
• Toggle Settings: When starting a conversation, you can toggle which actions are displayed in the transcript. In expanded view, hovering over an action in the side panel highlights the related utterance for easier traceability.

​

Observing GenerativeAgent’s Behavior

• Its thoughts during the conversation
• The Knowledge Base articles it references
• The API calls it makes
• Active task configuration
• Current reference variables
• Precise instruction parsing
• Function call context and parameters
• Execution state at each conversational turn

​

Mid-conversation experimentation

• Regenerate a response: Regenerate a GenerativeAgent’s reply using the latest draft settings to see how changes affect the outcome.
• Edit customer messages: Change a customer message and observe how the GenerativeAgent responds with the new conversation context.

​

Replaying Conversations

​

Next Steps