The Previewer in AI Console makes it easy to rapidly iterate on GenerativeAgent’s design and provides a quick tool to test GenerativeAgent’s capabilities.

​
Testing Draft Changes

When you initially configure GenerativeAgent, you’ll often find subtle ways to improve its performance. While you can always make changes, then deploy and test them in sandbox, it’s usually easier to try changes with Previewer. Previewer can use any changes across tasks and functions that you have in draft, allowing you to interact with GenerativeAgent using these temporary configurations.

Once you’re confident with a set of changes, you can deploy them into sandbox.

​
Using Live Preview

The Live Preview feature allows you to test changes in real-time during a conversation. You have the ability to:

  • Regenerate a response: For a given bot response, regenerate it using the latest state of the draft settings.
  • Send a different message: For a given customer message, change what is sent to see how GenerativeAgent would respond with that conversation context.

​
Previewer Environment

Choose the Environment that GenerativeAgent uses to test and preview a conversation with GenerativeAgent.

Choose between:

  • Draft
  • Sandbox
  • Production

​
Replaying Conversations

During testing and configuration, you may want to replay conversations while trying out changes or validating GenerativeAgent across new versions. In Previewer, you can save the conversation to replay it again in the future.

​
Advanced Settings

Use Previewer’s Advanced Settings to further test GenerativeAgent in the Previewer.

​
Test User Type

Use the test user data or reach out to an existing API Connection.

Test Users allow you to define a scenario and how your API would respond to an API Connection for that scenario. This allows you to try out different Tasks and iterate on tasks definitions or on Functions.

  1. API Connection: The Previewer test the conversation with mocked data defined by a Test User.
  2. External Endpoint: The Previewer uses external data from an existing API.

​
Task Name

Choose a specific Task to make GenerativeAgent handle it, instead of allowing GenerativeAgent choose a Task each time the conversation is started.

If you leave the task name blank, then GenerativeAgent will choose the Task by itself.

This way you can test:

  1. How GenerativeAgent handles a specific Task
  2. How GenerativeAgent chooses Tasks to perform a Function.

Task name is also an optional part of the body request in the GenerativeAgent API with /analyze.

Head to Improving Tasks to learn more about the use of Tasks.

​
Input Variables

Input Variables allow you to simulate how GenerativeAgent responds when it receives data from a calling application during a conversation.

Use Input Variables to test the use of:

  • Entities extracted from a previous system or API call
  • Relevant customer metadata
  • Conversation context, like a summary of previous interactions
  • Instructions on the next steps for a given task

Input variables can be submitted as key-value pairs in JSON format. For optimal configuration, reference the input variables directly in the task instructions to guide GenerativeAgent on how to interpret them

You can also simulate directly launching the customer into a specific task, instead of allowing GenerativeAgent to choose a task.

In a scenario where a IVR has already gathered information, you can ensure GenerativeAgent picks up from where the IVR left off.

​
Observing GenerativeAgent’s Behavior

Previewer gives you insight into the actions that GenerativeAgent is taking. This includes its thoughts during the conversation, the Knowledge Base articles it references, and the API calls it makes.

You can use this information to evaluate the performance of your tasks and functions, making appropriate changes when you want to alter its behavior.

​
Turn Inspector

Use the Turn Inspector to examine how instructions are processed within GenerativeAgent.

Inspect the state of the variables, tasks, and instructions in each turn of conversation within the Previewer.

Turn Inspector includes detailed visibility into:

  • Active Task Configuration
  • Current reference variables
  • Precise instruction parsing
  • Function call context and parameters
  • Execution state at each conversational turn

​
Next Steps

You may find one of the following sections helpful in advancing your integration:

Integrate GenerativeAgent

Safety and Troubleshooting

Go Live

Was this page helpful?

Test UsersFunctional Testing