Docs

Observability Platform

AI Application Insights

Insight rules

Insight rules are the criteria that the system uses when analyzing AI application sessions. They define the analysis objectives, evaluation logic, and output requirements. When analyzing a session, the system passes the session content as a variable into the system prompt configured in the rule, and uses a large language model (LLM) to generate structured insight results. Viewing insight rules helps you confirm whether the evaluation criteria and output content meet the current analysis objectives before you create an insight task. This prevents misinterpretation of results due to inconsistent understanding of the rules. This topic describes how to view insight rules.

Prerequisites

You have view permissions on the target workspace. For details, see Workspace-level IAM permission control.

Limits

Currently, you can only view system-preset rules (for example, session satisfaction). Creating, editing, or deleting insight rules is not supported yet.

View rules

In the left navigation pane, select AI Application Insights > Insight rule to go to the insight rules list and view the currently available rules.

Click the name of an insight rule or Details in the Actions column to view the rule details.

Settings
Previous versions
On the Settings tab, you can view the rule details.
Basic information: Includes the rule version, rule ID, data granularity, and output rules.
System prompt: Defines the role that the LLM plays during analysis, the tasks it must complete, the evaluation criteria, and the output requirements. It is the core basis on which the model generates insight results. Viewing this field helps you understand how the model analyzes sessions and determine whether the current rule matches your analysis objective. Because the system prompt may vary, the tags and evaluation rationales generated by the model can also differ accordingly.
Template variable: Defines the data to be analyzed that is passed into the system prompt. For example, session_messages represents the complete message list of the current session, including user input, agent responses, and tool calls. Viewing this field helps you confirm the actual scope of session data that the rule analyzes, and avoid misinterpreting analysis results due to a mismatch between the input data scope and your expectations.
Output definition: Defines the result type that the LLM returns, the tag values, and the evaluation criteria for each tag. Viewing this field helps you accurately understand the meaning of each tag and confirm whether the output meets subsequent screening and analysis requirements.
On the Historical versions tab, you can view the version history of the rule. This includes information such as version number, creator, and creation time. By default, the latest version is displayed and marked as "vN (Current version)".
tip
Version history is for viewing only. Version rollback is not supported yet.

Other operations

Click Create insight task in the Actions column. The system automatically associates the rule and creates an insight task. For details, see Creating insight tasks.

Last updated: 2026.09.21 10:36:58