1. Where can I view detailed user logs or transcripts for Copilot Studio Agents? Can I search by Conversation ID?
Activity Tab:
The Activity tab in Copilot Studio lets you review agent runs, but currently, it typically requires a Run ID (not Conversation ID) to search for specific logs.
Conversation ID Search:
As of early 2024, searching directly by Conversation ID is not natively supported in the Copilot Studio UI. You need the Run ID to locate a specific session.
Workaround:
If you have access to the underlying Dataverse tables (e.g., via Power Platform Admin Center or Advanced Find), you may be able to query for conversations using the Conversation ID.
Alternatively, export logs or use Power Platform analytics if enabled for your environment.
Permissions:
You may need additional permissions to access Dataverse tables or analytics.
2. Do I need to be an Administrator or Environment Maker to access conversation logs?
Required Role:
To view detailed activity logs and transcripts, you typically need to be an Environment Maker or System Administrator in the Power Platform environment where the agent is deployed.
Basic users or those with only agent authoring permissions may not have access to full logs or analytics.
Check your role:
Go to Power Platform Admin Center → Environments → [Your Environment] → Settings → Security roles.
3. How can I determine whether the ContentFiltered error was triggered by Responsible AI guardrails or another cause?
ContentFiltered Error:
This error is triggered when the agent’s response is blocked by Microsoft’s Responsible AI content filters (e.g., for sensitive, unsafe, or inappropriate content).
Diagnosing the Cause:
In the Activity tab, if you can access the run/transcript, look for system messages or flags indicating a safety block.
If you have access to the transcript or message history, review the user’s input and the agent’s attempted response at the time of the error.
If you cannot access the transcript, you may need to request elevated permissions or ask an admin to retrieve the relevant run.
Further Steps:
Review the user’s prompt for potentially sensitive or filtered content.
Check if your agent’s knowledge sources or plugins could have returned content that triggers filters.