Usage Tracking

Overview

Usage Tracking displays the operational metrics of the current dataset agent and provides query and management capabilities for dialog history, helping administrators understand how the agent is being used and how users interact with it.

Usage Guide

View Operational Metrics

The operational metrics module presents the core usage status of the current agent in a data-driven way, helping administrators fully understand its operational performance.

No.	Description
1	Toggle whether operational metrics are shown in expanded detail.
2	Switch the time range used to view operational metrics. This filter also applies to Dialog History.
3	View the number of questions asked to the current dataset agent, evaluations of answer quality, active users, and the number of users who asked questions. Question Count: reflects overall usage frequency and usage level. Active Users: reflects user stickiness and helps identify core user groups. Answer Feedback: comes from Answer Feedback and directly reflects answer accuracy and user satisfaction, making it an important basis for Q&A optimization. Asking Users: shows how many users participated in asking questions, reflecting the coverage of the agent.

View Dialog History

Each frontend question generates a backend dialog history record. You can filter dialog history by answer feedback result, dialog type, dialog status, exception status, full message ID or dialog content, time range, and user name.

In the dialog history list, the system automatically marks abnormal records that encountered issues during execution, helping operations staff quickly identify records that need attention.

Export Dialog History

Dialog history records support one-click export, including Q&A status, whether the process was abnormal, and exception descriptions, making it easier for operations staff to perform offline analysis and troubleshooting.

Copy Message ID

Click the button to copy the message ID. The copied message ID can be provided to Guandata for troubleshooting.

View Langfuse Trace Records

Click Trace to jump to the Langfuse platform and view detailed call chain information for troubleshooting.

Manage Dialog History

Click Dialog History to display conversation records. The visualization shown in backend dialog history stays consistent with the frontend experience, and you can manage records from the detail page.

Q&A Dialog History

When the dialog type is Q&A, the Dialog History page supports operations such as viewing retrieved knowledge, viewing notification history, adding to Q&A Sample, viewing SQL, copying the message ID, and managing likes/dislikes.

No.	Feature	Description
1	Show / Hide Analysis Parameters	The query result area supports showing or hiding analysis parameters to help users understand the data query logic.For details, see Show / Hide Analysis Parameters.
2	Enlarge	If the current page cannot display all chart content, click the enlarge button to view the chart more clearly.
3	Export	Export the data result. Both Export Excel and Export Table Data are supported. For details, see Export Data.
4	View SQL	Click SQL to view the SQL used for the current Q&A. Click SQL Explanation to view the execution logic, filter conditions, and data relationships, or click the copy button to copy the current SQL.
5	View Retrieved Knowledge	Displays the Table Knowledge, Business Knowledge, and Q&A Sample sent to the model during Query. There are two entry points for this feature in Q&A dialog history: one in the upper-right corner of the dialog detail page after clicking Dialog History, and another in the right-side action list under Usage Tracking > Dialog History. For details, see View Retrieved Knowledge The new Retrieved Knowledge feature is not compatible with the old Operational Logs view. When upgrading, note the following: For private deployments of datamind, if historical issues need to be investigated after the upgrade, copy the dialog ID and search it in Langfuse. For SaaS environments, complete issue investigation before upgrading to avoid losing the ability to view retrieved knowledge in historical dialogs.
6	View Notification History	If Email Push was used in frontend Q&A, you can view notification records in Notification History.
7	Add to Q&A Sample	Automatically fills the current question into the description and the current SQL into the query SQL field. After correcting the SQL, users can click Add to Q&A Sample to quickly add the Q&A pair to Q&A Sample knowledge. There are two entry points for this action: one in the upper-right corner of the dialog detail page after clicking Dialog History, and another in the right-side action list under Usage Tracking > Dialog History. For details, see Add to Q&A Sample.
8	Like/Dislike Management	For Q&A entries that users did not evaluate in the frontend, administrators can mark them with Like or Dislike when reviewing dialog history. Dislike entries can include feedback text. For details, see Manage Like/Dislike Records.
9	Copy Message ID	Click the button to the left of the message ID to copy it for troubleshooting.

Attribution Dialog History

When the dialog type is Attribution, the Dialog History page supports viewing notification history, copying SQL, adding to Q&A Sample, viewing retrieved knowledge, and like/dislike operations.

No.	Feature	Description
1	Insight Analysis Process	Click Completed Insight Analysis to expand Analysis Process. In the Analysis Process, find Data Query and click it to open the View Calculation Details page, where you can copy SQL, add to Q&A Sample, and view retrieved knowledge. Copy SQL: copies the SQL directly for data warehouse queries. Add to Q&A Sample: fills in the current question and current SQL automatically. After correction, click Add to Q&A Sample to store it in Q&A Sample. For details, see Add to Q&A Sample. Retrieved Knowledge: shows the Table Knowledge, Business Knowledge, and Q&A Sample sent to the model during the query. For details, see View Retrieved Knowledge.
2	Like/Dislike Management	For records not evaluated in the frontend, administrators can add Like or Dislike tags and provide feedback when needed. For details, see Manage Like/Dislike Records.
3	Copy Message ID	Click the button to the left of the message ID to copy it for troubleshooting.

Manage Like/Dislike Records

For records that users did not evaluate in the frontend, administrators can review dialog history and mark problematic answers with Like or Dislike. When marking a Dislike, feedback text can be entered.

Dislike Workflow

1. Click Dislike, enter the issue description, and click Submit. Quick-fill inputs are supported to help define unclear or problematic answers.

   

2. After a user submits a dislike, the current issue is marked as To Be Optimized. After the knowledge base administrator finishes the optimization, the result can be labeled.

   • No Action Needed: the actual behavior matches expectations, so no knowledge base optimization is required.
   • Processed: optimization has been completed, and frontend users can try again.
   

3. After processing is complete, the knowledge base administrator fills in the processing notes and submits them. The system records the result.

   

4. (Optional) If the recorded result is incorrect after changes, click Edit Processing to modify it.

   

Add to Q&A Sample

After clicking this action, the current question is automatically filled into the description field, and the current SQL is filled into the query SQL field. After the SQL is corrected and passes validation, click Add to Q&A Sample to quickly add the Q&A pair into Q&A Sample knowledge.

The system checks and prompts for the current SQL syntax type. For extracted datasets, the syntax type is displayed as Spark. If mixed SQL across different direct-connect datasets is detected, the system reports that cross-database queries are not supported.

View Retrieved Knowledge

Retrieved Knowledge displays the Table Knowledge, Business Knowledge, and Q&A Sample that were sent to the model during the query.

• Dataset Knowledge: lists the table information and corresponding enumerated fields recalled in this calculation.

  

• Business Knowledge: shows the business rules or knowledge items associated with the current calculation. Users can edit the knowledge or add new Business Knowledge.

  

• Q&A Sample: shows the historically corrected SQL recalled by the system during the current calculation.

  

Note

• The knowledge shown here is the content recalled for this specific Q&A run. If the knowledge was updated later by you or your colleagues, clicking Edit will show the latest version from the knowledge base rather than the old content shown in the list.
• If a piece of Business Knowledge or Q&A Sample has already been deleted from the knowledge base, the Edit entry becomes disabled to preserve knowledge accuracy.