This quickstart shows you how to use STAROps Intelligent Conversation to query observable data in natural language.
What you will accomplish
After you complete this quickstart, you will be able to do the following:
Start an Intelligent Conversation from the STAROps console.
Ask a digital employee questions in natural language and obtain query results for observable data.
Step 1: Open the Intelligent Conversation interface
Log on to the STAROps console.
In the sidebar, click New Conversation to open the Intelligent Conversation interface.
Step 2: Select a digital employee
At the top of the conversation interface, select the digital employee that you want to converse with. Different digital employees have different capability configurations and data access scopes.
If only one digital employee is available, the system selects it automatically.
Step 3: Select a data scope and a model
Before you start the conversation, you can specify the AI model and the data query scope for the current conversation as needed.
Set a data scope (optional)
At the top of the conversation interface, click Add Data Scope.
In the selector that appears, select a workspace (CMS2.0 workspace) or a Simple Log Service (SLS) project.
Select an AI model
At the bottom-right of the conversation interface, find the model selection drop-down list.
Click the drop-down list to view the available models:
Auto: uses the default model configured on the server and does not perform automatic switchover between models. This is the default option.
Qwen: suitable for general question answering and Chinese content generation.
GLM: suitable for technical analysis and code comprehension.
DeepSeek V4-Flash: suitable for lightweight tasks that require fast responses.
DeepSeek V4-Pro: suitable for complex tasks that require in-depth analysis.
Click the target model to complete the selection.
The Data Scope selector replaces the previous workspace selector.
Step 4: Start the conversation
In the input box, describe your query in natural language. The following table lists common examples of questions.
Scenario | Example question |
Data query | Show the top 5 hosts with the highest CPU usage over the past 24 hours. |
Outlier detection | What alerts occurred in the last hour? |
Fault diagnosis | Analyze the cause of the response latency of |
After you enter the question in the input box, press Enter or click Send, and then wait for the digital employee to return the result.
Step 5: View the results
The digital employee automatically invokes the relevant tools to query data and presents the results in the conversation interface as text descriptions, data tables, or metric trend charts.
If you need further analysis, ask follow-up questions in the conversation. For example:
"What is the root cause of this alert?"
"Compare the data with the data from the same period last week."
Next steps
After you finish the basic Intelligent Conversation experience, you can explore the following capabilities:
@entity reference: In the dialog box of the Cloud Monitor entry point, type
@to reference entities in the workspace, such as Service, Node, and Pod, so that the AI can locate the analysis target more precisely. For more information, see the intelligent assistant in Cloud Monitor.'/' invocation: Digital employees also provide the '/' capability. You can use '/' to invoke commands such as
/Skill,/Tool,/clear, and/compact.Long-running tasks: For scheduled routine inspections or automated O&M, see the quickstart for long-running tasks.
Default rules: Writing default rules for a digital employee significantly improves the conversation quality. For more information, see Configure default rules.
FAQ
An operation failed due to insufficient permissions
Check the following possible causes:
Check whether the digital employee has the relevant permissions.
Check whether observable data is ingested into the workspace. If the data source configuration is incomplete, the digital employee may be unable to process the query.
Check whether the RAM role associated with the digital employee has read permissions on the corresponding data source.
The digital employee returns inaccurate results
You can improve the accuracy of answers in the following ways:
Be as specific as possible in your questions. Include context such as the time range, the target resource, and the metric name. For example, change "View CPU" to "Show the top 5 pods with the highest CPU usage in
prod-clusterover the past 24 hours."Use the @entity reference feature to specify the analysis target precisely. Type
@in the dialog box, select the target entity, and then send the message.Configure default rules for the digital employee to specify the requirements for the output format and the analysis depth. For more information, see Configure default rules.