Skip to main content
Admin › Evaluation › Feedback
The response evaluations users leave in chat and Arena accumulate as feedback records.
  • Each record shows the evaluated model and its result (win/loss), and can be shared to a community leaderboard.
The like/dislike buttons on responses appear only when Enable Message Rating under Admin › Settings › General › Features is on (on by default). Turning it off stops new evaluations from accumulating, so no new data reaches this page or the leaderboard (existing records are kept). The rating buttons are also hidden in Temporary Chat.
Feedback records table

Evaluation › Feedback — feedback records table

​
Feedback Records

Each row is one evaluation. Clicking the View Trace icon on the right of a row opens Evaluation › Tracing in a new tab, automatically searching by that evaluation’s Chat ID. The icon appears only on evaluations that also recorded a Chat ID or Message ID, and the Tracing tab does not open in environments where Tracing is disabled.
  • The ⋯ menu (More) contains only Delete. It deletes immediately without a confirmation dialog and cannot be undone, and an admin can also delete evaluations left by other users.
  • The list is paginated.
Clicking a row opens the Feedback Details dialog. It shows the ID, Message ID, and Chat ID, a View Trace button, the user, the result, and the Evaluated Model; Compared Models, Reason, Comment, and Tags appear only when they have values. Reason, comment, and tags are what the user wrote in the rating dialog after clicking like or dislike. Updated at in the list is a relative time, while Updated in the dialog shows the exact timestamp (YYYY-MM-DD HH:mm:ss).
Feedback Details dialog

The Feedback Details dialog that opens when you click a row

​
Export and Share

The exported JSON file also contains a full conversation snapshot for each evaluation. It carries far more than a community share does, so always check the file’s contents before passing it outside your organization.
Clicking Share to Open WebUI Community opens the external community leaderboard in a new tab and sends your feedback records to it (nothing is sent if the pop-up is blocked). What is sent includes ratings, model IDs, compared models, tags, the rating reason and comment text, and Chat/Message IDs; full conversations and user names or email addresses are not included. The reason and comment do carry sentences the user typed verbatim, so check your organization’s policy before sharing.

​
Use Cases

  1. Identify models/agents with many losses in the feedback records
  2. Check the evaluation rankings across models in the Leaderboard
  3. Analyze the cause of low-rated responses with Tracing
  4. After improving the prompt and Knowledge Base, check the score change in Auto-Evaluations

Evaluation

Full overview and guide to evaluation methods

Auto-Evaluations

Automatic quality scoring by a judge LLM

Tracing

Follow the trace of an evaluated response