Configuration and administration
Logs and user feedback
Inspect execution records and user ratings to diagnose problems and improve answers.
Guide · Module: agent · Users · Deployment: Not specified · Confirmed product versions: Not specified · Source: Existing documentation
Audit logs record how each question was executed, while user feedback records ratings of answers. Use both to diagnose problems and improve the agent's responses.
Audit logs
Open the Audit logs tab. Its description says it shows all users' questions and execution records for the current agent. The list refreshes automatically every 15 seconds.
List and filters
| Column | Contents |
|---|---|
| Question | The user's original question; shows “Question text not recorded” when absent |
| User | The person who asked |
| Source | Entry point, such as web conversation, Feishu, Slack, or API key |
| Time | When the question was asked |
| Execution status | Outcome status for the execution |
| User rating | Positive / negative / unrated |
Filters include question search with a Search button, user (defaults to All users), execution status, start date, and end date. Page sizes are 10 / 20 / 50.
Execution details
Click a record to open “Execution details” on the right. The page reminds you: Only saved records are shown; a recorded reply does not mean the execution completed successfully.
Details include:
- The question and execution status;
- Each message's role, time, and content;
- Expandable tool calls under each message, such as semantic retrieval, SQL execution, chart generation, and Python execution;
- Error information, including original JSON;
- The user's rating of the answer.
“Search this execution” locates keywords throughout the turn's records. “Clear search” resets it. The panel shows “No matching records” if there are none, or “Matching records: N” otherwise.
What to investigate
- Wrong business definition: when a user says numbers are wrong, expand the tool calls to inspect the executed SQL and tables used. Decide whether to improve field descriptions or business documents.
- Cannot answer: check whether table retrieval failed. Common causes include a source not attached to the agent or a policy blocking access.
- Slow answers: identify the slow step to distinguish data source latency from model latency.
User feedback
Open the User feedback tab, whose description says submitted feedback is handled here centrally. Users can add a comment when giving an answer a thumbs-up or thumbs-down.
List and filters
| Filter | Options |
|---|---|
| Status | Pending / Resolved / All |
| Feedback type | All feedback / Thumbs-up / Thumbs-down |
Each item shows the rating, the user's comment (or a summary of the rated message if no comment was provided), avatar and nickname, conversation name (“Unnamed conversation” if absent), and relative time.
Process feedback
- Click Mark resolved to archive feedback. The button changes to Reopen, which returns it to pending.
- Click the whole row to open the associated conversation and inspect the original context.
Use feedback to improve answers
Feedback is an error report from the user's perspective. Group negative comments by cause; they usually point to three areas:
| Typical feedback | Likely missing information | Where to improve it |
|---|---|---|
| “This number is wrong” / “That is not our definition” | Unclear field descriptions or business definitions | Table and field descriptions · Business documents |
| “The answer is too wordy / too brief” | Undefined analysis preferences | Basic settings and knowledge preferences |
| “It did not follow our analysis method” | A missing skill | Analysis skills |
Key constraints
| Item | Value or behavior |
|---|---|
| Log scope | All users' questions for the current agent, across entry points |
| Log refresh | List refreshes every 15 seconds |
| Log details | Only saved records are shown; a recorded reply does not mean full execution success |
| Rerun or export logs | Neither is available in the interface; logs are view-only |
| Feedback status | Pending and Resolved; can switch in either direction |
| Feedback visibility | Available to administrators of the agent, with links to original conversations for context |
Troubleshooting
| Symptom | How to identify it | What to do |
|---|---|---|
| Empty log list | “No audit logs” appears | Confirm the agent has actually been used; try another agent or widen the date range |
| Question shows “Question text not recorded” | Original question text was not saved | Inspect message content in Execution details instead |
| Cannot find a question | Search returns nothing | Shorten keywords, clear user/status filters, or widen the date range |
| Details do not show the complete process | Reminder says only saved records are displayed | This is expected; details reflect only the persisted portion |
| Empty feedback list | “No user feedback” appears | Users have not rated answers yet; encourage them to provide feedback |
| Feedback cannot open its conversation | Conversation was deleted or access is denied | Assess the feedback directly and ask the user about the original question if necessary |
Next steps
- Turn negative feedback into clear definitions: Business documents
- Capture repeated analysis methods: Analysis skills
- Check that capability switches match expectations: Capability switches explained