Models
A model configuration stores the service source, model name, endpoint URL, and parameters. Evaluators use models for scoring, and large language model evaluations reuse the same type of configuration.
For users who maintain model services. Verification baseline: the project's current implementation as of 2026-10-30. Entry points in deployed environments may vary by version and permissions.
Search keywords: scoring model, model under test, AE Agent, OpenAI-compatible, Anthropic, Gemini, JSON scoring.
1. Create an AE Agent model
- Go to Implementation → Model and click Create model.
- Select AE Agent (A2A 1.0) as the Provider and enter a name for the model configuration.
- Wait for the catalog to load, or refresh the model list, then select a model from AE Agent Model.
- In a TE SSO environment, the current login access token is used automatically. You do not need to create or select credentials manually. The environment generates the model URL automatically and displays it as read-only.
- After saving, run Test connectivity from the list. Confirm success before using the model in an evaluator or task.
The production URL defaults to the current page's origin plus /agent. Local development URLs depend on the development environment configuration. If the catalog is empty or fails to load, check your login status, environment URL, and error messages. An empty catalog does not mean you lack permission to create evaluators.
2. Create other models
- Select OpenAI-compatible, Anthropic, or Gemini, then enter the configuration name, model name, and model URL.
- Select available credentials according to the service requirements. The service maintainer should provide the endpoint URL, model identifier, and authentication method.
- Set Temperature and Top P under Advanced Settings as needed, then save and test connectivity. If the service requires fixed parameter values, correct them according to the returned error.
Azure OpenAI is not available when creating a model in this version. Existing Azure configurations can still be edited. A provider's presence in type definitions does not mean it is available in the creation UI.
3. Reduce scoring format errors
For OpenAI-compatible and existing Azure configurations that support JSON Schema response_format, enable Enforce JSON scoring format to constrain Judge output to Score and Reason. The service must actually support this protocol. If it does not, disable the option and adjust the model or Prompt. This option does not replace the scoring criteria themselves.
4. Edit and verify
- Edit the configuration from the model list, save your changes, and test connectivity again. If saving fails, retain the form content, correct the issue, and submit again.
- Before using the model in an evaluator, check its enabled status, credentials, and project visibility.
- Changes to model configurations do not rewrite existing run snapshots. Use the configuration saved in the Run when analyzing historical results.
When connectivity errors occur, first distinguish authentication failures, URL or network issues, unsupported model identifiers, and incompatible parameters. If failures persist, provide the administrator with the error code, request ID, and time of occurrence. Do not send keys.
5. Next steps and related pages
- Prepare an identity for services other than AE: Credentials.
- Select a Judge model: Evaluators.
- Select a large language model to evaluate: Create evaluation tasks and map fields.
Return to the manual home page:

