Execution trace helps you understand how an AI-generated reply was produced in Inbox. It shows a read-only breakdown of the logic behind a historical AI message, including the detected intent, referenced knowledge, and playbook execution results.
This helps you review past AI replies more clearly, especially when a response feels incorrect, incomplete, or unexpected. Instead of guessing what happened, you can inspect how the AI understood the message, what information it used, and which playbook steps were evaluated.
Before you start
Execution Trace is available only for AI-generated messages created after this feature is launched in Inbox. It is a read-only view, so you can review how a reply was generated, but you cannot edit prompts or agent configuration from this view.
When to use Execution trace
Execution trace can help you:
- Review why the AI responded in a certain way
- Check the intent the agent understood from the end customer’s message
- See which Knowledge Base content was referenced
- Review which playbook steps were executed or failed
- See which external tools were called, such as Shopify actions or custom APIs
- Better understand how the AI generated a reply by reviewing available execution details
Note:
Execution Trace is a read-only diagnostic view for AI-generated messages created after the feature is launched. Execution Trace data is available for 90 days only. In this initial release, it focuses on trace visibility and does not support updating prompts or agent configuration from this view.
Accessing the Execution trace
You can follow the steps below to access the Execution trace
- Click on the
icon to go to Inbox - Open a conversation that contains a historical AI-generated message.
- Click “Execution trace” below the AI message bubble.
- The Execution trace drawer opens on the right side
Review how the AI understood the message
In the “Understand” section, you can review how the AI interpreted the customer message.
This section shows:
- The detected Intent
- The collected intent information for that reply
Use this section to check whether the AI classified the message correctly before generating a response.
Review what knowledge the AI referenced
In the Recall section, you can review the knowledge used to generate the reply.
This section shows:
- The referenced Knowledge Base article title
Use this section to verify whether the AI pulled the right source content for the reply.
Review how the playbook was evaluated
In the “Act” section, you can review how the AI evaluated the playbook for that message.
This section shows the Playbook Execution list, so you can see which steps were evaluated for the historical reply.
This is useful when you want to understand whether a playbook step was completed or failed during execution.
Understand execution statuses
For historical AI messages, Execution trace shows resolved static statuses only.
You may see these statuses:
- Executed: shown with a green check
- Failed: shown with a red X
Loading states and spinners are not shown for historical traces.