The inference capability of OAgent is provided by external large language models; the platform does not have a built-in large model. Administrators must add a model service on the Model Configuration page for sessions to function properly.
OAgent calls external large model APIs through standard protocols. The following protocols are supported; choose one when integrating:
- OpenAI-compatible interface (Chat Completions):
/v1/chat/completionsformat, compatible with OpenAI-compatible endpoints of mainstream model services such as Qwen, DeepSeek, and Alibaba BaiLian. - OpenAI Responses API:
/v1/responsesformat, an interface launched by OpenAI that supports built-in tool invocation and state management.
Click Model Configuration in the left navigation bar to go to the model configuration page. The page is divided into two tabs: Dialogue Model and Vector Model:
Dialogue Model: Configure models for dialogue inference and tool invocation. You can add multiple model services by service name.
Vector Model: Configure vector models for knowledge base retrieval and memory indexing. The platform supports saving only one configuration.
Dialogue model
Manage model services by their service names in the configuration: Each service name corresponds to a card, which can configure multiple models and share the same API address and key. The card displays the service name, API address, and a summary of configured models (for example, qwen3.5-plus, deepseek-v4-flash, qwen-72b-chat, etc.), and has status labels Include Default Model and Key Configured.
Add a model service
On the Dialogue Model tab, click Add Model Service.
Enter the following parameters:
ParameterDescriptionService Name The internal service name identifier after saving; supports Chinese characters, uppercase and lowercase letters, numbers, and hyphens. API Address The access address of the model service, for example, https://dashscope.aliyuncs.com/compatible-mode/v1。`API Key The access key of the model service. In the Add Model input box, enter the model name and press Enter to add it (optional). If the API key has been entered, the system will attempt to obtain the list of models supported by this service for selection; if the model list cannot be obtained, you can directly enter the model name and press Enter to add it.
Click OK to save the configuration. At least one model must be added under the service; otherwise, it cannot be saved.
Notice
The API key cannot be modified after being saved. The saved model name also cannot be modified. If you need to change it, delete it and add it again.
Test a model
Click Test on the model card to open a full capability test, verifying whether the API address and key are usable, and whether the model's tool invocation capabilities meet the requirements of operational diagnostic scenarios.
On the model service page, click Evaluation Record to view the execution progress and historical reports of the full capability test. Each record includes the model and its corresponding model service, target level, score, capability conclusion, status (Passed / Failed), and creation time; clicking View Report opens the detailed report for that test, allowing you to compare the performance across multiple tests and models.
For report viewing and result interpretation of evaluation records, see Model scoring.
Set a default model
On the model service page, you can set a model as the default model. After saving, the Include Default Model label will appear on the corresponding card. After selecting a default model, click OK to save the change. New sessions will use this model by default, and users can switch models within a session. For details, see Switch dialogue models.
Delete a model service
On the corresponding Model Service page, click Delete Model Service and confirm to delete the service and all models configured under it. If you need to use it again after deletion, you must add it back.
Usage suggestions
Configure at least one dialogue model that supports tool invocation; otherwise, skills that rely on tools for inspection and diagnostics cannot be executed.
It is recommended to configure a model with stronger inference capabilities for root cause analysis scenarios.
Before using the knowledge base retrieval capability, complete the vector model configuration on the Vector Model tab.
Vector model
Only one set of vector models can be configured on the Vector Model tab, which is used for knowledge base retrieval and memory indexing.
Enter the API endpoint.
Enter the API key (optional).
Enter the Model name.
Click Save to complete the configuration. After configuration, you can click Test connection to verify if the vector model is available.
Note
An API key cannot be modified after being saved. If you need to replace it, first click Delete Configuration to remove the current configuration and then add a new one. After the configuration is deleted, knowledge base retrieval and memory indexing will be unavailable.