Reviewing AI Assistant chat sessions
This guide helps helpdesk staff review the conversations users have had with the Waldur AI Assistant — to follow up on a support request, audit flagged content, or understand how the assistant is being used.
Note
The review pages are available to staff and support users when the AI Assistant is enabled. For turning the assistant on and configuring it, see AI Assistant configuration.
Accessing the chat session logs
- Select Support from the left-hand menu.
- Click Logs from the top menu.
- Open the AI assistant logs tab.
A table opens listing every chat session (thread). Each row shows:
- The user and their organization.
- The thread name (auto-generated from the conversation).
- The number of messages and the token usage of the session.
- Whether the session is flagged, its max severity, and whether the user left feedback.
- When the session was created and last modified.
Finding a specific session
The table can be narrowed down in two ways.
Search by keyword or phrase
Use the search box above the table to find sessions by their content. The search covers:
- The conversation itself — both the user's questions and the assistant's replies.
- The thread name.
- The user — username, name, or email address.
The search understands natural phrasing, so you can type a few keywords
(resize openstack volume) or wrap an exact phrase in quotes
("quota exceeded") to match it as a whole.
Tip
Content search matches word stems, so resizing also finds sessions that
mention resize. If a search returns too many results, add another keyword
or switch to an exact quoted phrase.
Filter by attribute
Use the column filters to focus on a subset of sessions:
- User — sessions belonging to a specific person.
- Flagged — sessions where the input guards flagged a message.
- Feedback — sessions where the user rated a reply.
- Max severity — the highest severity detected in the session
(
Critical,High,Medium,Low, orNone).
Filters and the search box combine, so you can, for example, search for a keyword and restrict the results to flagged sessions.
Reviewing a conversation
Expand a row to read the full transcript of the session. For each turn you can see the user's message, the assistant's response, and any warning raised by the input guards (for example, redacted personal data or a detected prompt-injection attempt).
Warning
Chat sessions may contain personal data. Access is restricted to staff and support users and every review is auditable — only open sessions you have a legitimate support reason to review.
Related
- AI Assistant configuration — enabling the assistant, access roles, token limits, and session retention.
- Service desk interactions — managing support tickets raised by users.