Following this topic, you can complete an end-to-end workflow in the shortest path: Log in → Configure models and connect to products (order is flexible) → Initiate the first cluster health inspection. After completing this section, you will obtain your first evidence-based inspection conclusion, which you can then apply to more scenarios such as alert analysis and SQL diagnostics.
Prerequisites
Before you begin, ensure the following conditions are met:
You have logged in to the OAgent console.
You have obtained the API endpoint and API key of a large language model service (such as Qwen or DeepSeek). Skills like session reasoning and tool scheduling rely on models; inspections cannot be performed if no model is configured.
You have obtained the service address and login credentials of the OCP instance to which the target OceanBase cluster belongs (the account must have view permissions for the target cluster and tenant). For diagnostic capabilities such as slow SQL analysis, ASH, and root cause analysis, you also need to obtain the service address and HTTP Basic authentication information of OAS.
Network connectivity exists between the OAgent Gateway and the service addresses of OCP/OAS.
To quickly understand the positioning and capabilities of OAgent, see What is OAgent. For the interface structure after login, see Interface navigation and overview page.
Procedure
Perform the operations outlined below. Configuring model services and connecting to products is not sequential; you can proceed according to your actual situation.
Configure model services
Skills such as inspection and diagnostics rely on models for reasoning and tool scheduling. Before using them for the first time, you must configure at least one dialogue model that supports tool invocation.
In the left navigation bar, click Model Configuration to go to the Dialogue Model tab, then click Add Model Service.
Enter the Service Name, API Endpoint, and API Key, add the names of models available under this service, and click OK to save the configuration.
Mark the model to be used in inspections as the default model.
It is recommended to configure both a lightweight model with fast response and a model with stronger reasoning capability, for information query and root cause analysis scenarios respectively. For complete parameter and connectivity testing, see Manage model services.
Notice
The API key cannot be modified after saving. To change the key, delete the model service and add it again.
Connect to OCP/OAS product connections
The product connection determines the scope of clusters and tenants accessible to the Agent and injects cluster information into the session context as needed.
At the bottom of the left navigation bar, click More to go to the Product Management page, then click Create Product.
Enter the Product Name (for example, "Production Environment OB"), OCP Address, and OCP Account Password. For diagnostic capabilities, also provide the OAS Address and HTTP Basic Credentials.
Click OK, and OAgent automatically pulls the cluster and tenant information under this product. For detailed field descriptions, see Create a product connection.
Initiate the first cluster health check
After connecting the model to the product, you can start the first diagnostic link: the cluster health check.
In the left navigation bar, click Create Dialog, enter
@in the input box, and select the newly configured product and target cluster. For details, see Use @ to mention the analysis object.Enter the inspection request. For example:
Help me inspect this cluster.You can also directly use the "Cluster Health Check" shortcut entry below the input box.
Press Enter to send it. Wait for the Agent to query multi-dimensional data such as the cluster's basic information, resource usage, active alerts, and slow/top SQL overview.
View the inspection conclusion.
The conclusion typically includes a problem overview, root cause analysis, key evidence, and prioritized remediation suggestions. Expand Tool Output to verify the original metrics cited in the conclusion. For details, see View the tool invocation process. For the complete inspection process and applicable scenarios, see Cluster health check.
Note
If the Agent indicates that a capability is unavailable, it is likely because the model is not configured or the corresponding skill/tool is disabled. Please return to the Model Configuration or Skill/Tool page to confirm. For details, see Skills overview.
Export reports and consolidate results
After the inspection is completed, you can consolidate the conclusion into a report for delivery and archiving.
In the session where the inspection conclusion has been generated, enter "Export this inspection conclusion as HTML".
The Agent assembles the report based on the obtained tool results and analysis conclusions, and provides options to download it in HTML format or open it directly in a browser. For details, see Export a report.
Time-series and proportion data in the report is presented as line charts, bar charts, pie charts, and tables. For details, see Data visualization.
What to do next
After completing the first inspection, you can expand as needed:
Schedule the inspection as a daily/weekly recurring task. For details, see Create a scheduled task.
Drill down in the same session to analyze the root causes of slow SQL, execution plans, or alerts. For details, see Slow SQL, TopSQL, and abnormal SQL analysis, Alert query and handling guidelines, and Root cause analysis of abnormal events.
Enable your team to use OAgent directly in DingTalk, Lark, WeCom, or Telegram workgroups. For details, see Multi-channel access.
Evaluate model performance and optimize the default model. For details, see Model scoring and Usage statistics.