Audit
The Audit page lets the Customer Super Admin review Agent usage across the organisation. Select a user and session to inspect the conversation, its timing, and the recorded steps associated with each response.
Open a session
- Open Audit under Transparency in the side menu.
- Search for and select a user.
- Select one of the user’s sessions.
- Review the Conversation and Audit Trace panels.
Use the session search and date fields to narrow the list. Relative times and event timestamps are displayed in your local time zone.
The Refresh button clears the current selection and reloads the latest users and sessions.
Page layout
| Panel | What it shows |
|---|---|
| Users | Users in your organisation who have Agent activity. |
| Sessions | The selected user’s recent conversations. |
| Conversation | User prompts and Agent responses in message order. |
| Audit Trace | The recorded progress, status, timing, and data-use information for each Agent run. |
Protect personal information
Lock Personal Information is enabled by default. While it is enabled, recognised personal identifiers are replaced in both the conversation and audit trace.
Protected information can include:
- People’s names.
- Email addresses and telephone numbers.
- Account and customer identifiers.
- National, tax, health, and social-security identifiers used in the United States, United Kingdom, European Union, Australia, and South Africa.
Business context such as monetary amounts, dates, percentages, roles, and departments remains visible where it is not itself a personal identifier.
To inspect authorised original values, turn Lock Personal Information off. The open session reloads automatically. Only unmask information when it is necessary for your review and your organisation’s policy permits it.
Sessions that used Mask to AI may display numbered placeholders while the lock is on. Where an original value is available and you are authorised to view it, unlocking personal information restores that value in the conversation.
Read the audit trace
The trace helps answer practical questions such as:
- Which datasets were selected?
- Did the request complete successfully?
- How long did it take?
- Which data actions contributed to the result?
- Was a governance rule applied?
Each run shows its start time, duration, status, and recorded activities. Expand a run to inspect the available input, output, timing, and error details.
Tree and Graph
- Tree presents runs as expandable rows and is best for detailed review.
- Graph presents the same activity visually and is useful for understanding the sequence at a glance.
Select a graph item to inspect its details.
Reasoning and User Feedback
- Reasoning shows the detailed trace available for eligible Reasoning sessions.
- User Feedback shows the concise progress updates associated with streamed responses.
These views record observable workflow activity; they do not display private model reasoning.
Exposure level
Responses can be labelled Full, Limited, or Advisory only. This identifies the LLM exposure level applied to that response. Use it when checking whether the detail in an answer was appropriate for the selected data.
Troubleshooting
- A new session is missing: click Refresh and confirm the date range.
- A conversation remains masked: confirm that the lock is off and that the original value is available for that session.
- The trace is empty: confirm that the session has a completed Agent response. Failed or interrupted requests may contain less detail.
- You cannot open Audit: the page is restricted to the Customer Super Admin.