Skip to main content

AI Model Management

AI Model Management lets administrators connect and maintain custom large language model services for NineData AI features. Use it when you want Text2SQL, ChatDBA, knowledge-base vectorization, or code generation to run on an approved model service.

What It Does​

  • Connect custom large language model services
  • Switch between built-in and custom models
  • Provide a shared model configuration for Text2SQL, ChatDBA, knowledge-base vectorization, and code generation

Before you begin​

  • Sign in to NineData with a system administrator account.
  • Create an API key from the target model provider.
  • Confirm that the model endpoint is reachable from the NineData environment.
  • Confirm the model type, request URL, API key, and underlying model name before you add the model.

Configure a Custom AI Model​

  1. Sign in to the NineData Console.

  2. From the left sidebar, go to System > AI Model.
  3. Click Add Model.
  4. Enter the model name, model type, request URL, API key, and underlying model.
  5. Click Save and enable the model configuration to save the configuration.

Enable or Disable a Custom Model​

  1. Sign in to the NineData Console.

  2. Open the AI Model list.

  3. In the action column for the target model:

    • Click Enable to enable the model.

    • Click Disable to disable the model.

      tip

      Only one custom model can be enabled at a time. If all custom models are disabled, NineData falls back to the built-in model.

Edit or Delete a Model​

  1. Sign in to the NineData Console.

  2. Find the target model in the list.
  3. Open the more-actions menu in the action column for the target model.
  4. Choose Edit or Delete for the target action.

Result​

Once enabled, the custom model is available to the AI features that use the shared NineData model configuration. After you switch models, test the key AI workflows that depend on it, such as Text2SQL, ChatDBA, knowledge-base retrieval, or code generation.

If model responses are slow, unstable, or lower quality than expected, check the model service availability, network connectivity, API key permissions, and model name before changing customer-facing AI workflows.