AI providers update and retire their models over time. Celigo displays lifecycle labels for models approaching retirement or no longer supported, helping you identify affected AI agents and guardrails and switch to a supported model before their runs are interrupted.
Anyone who can view AI agents, guardrails, or an integration sees the labels and icons. To switch a model, you need Manage or Administrator access to the AI agent or guardrail. Users with access to specific integrations see the Model attention panel only for those integrations.
Understand model lifecycle statuses
When you create or edit an AI agent or guardrail, each model in AI model selection has one of the following statuses:
| Status | Description |
|---|---|
| Supported | The model is available for use and doesn't require any action. |
| Migration recommended | The provider plans to retire the model, but no date is set yet, or the date is more than 90 days away. You can keep using it. Plan to switch to a newer one. |
| Retiring on date | The model retires on the date shown, which is 90 days away or less. Starting on that date, AI agents and guardrails that use it can't run or be saved. Switch to a supported model before then. The Model attention panel shows this status as Retiring soon. |
| Retired | The model has reached its retirement date and can no longer be used. |
| Unavailable | Celigo has turned off the model, for example because of a provider incident. |
| Not supported | The model isn't offered for this type of AI agent or guardrail (for example, a BYOK-only model on a Celigo AI agent), or Celigo no longer lists it. |
Note: Retired, unavailable, and unsupported models aren't available for new AI agents or guardrails. Existing AI agents and guardrails retain their selected models, but you must select a supported model before you can save changes or run them.
Lifecycle labels apply to models used with Celigo AI and Bring your own key (BYOK). If you use a BYOK model that isn't listed in AI model selection, Celigo doesn't track its lifecycle. Check your AI provider's deprecation notices for updates.
Identify models that need attention
Celigo displays warning and error indicators in the following locations:
- AI model selection - Lifecycle labels appear next to models approaching retirement. If an existing AI agent or guardrail uses an affected model, a message below the field identifies its status and recommends a replacement when available.
- AI agents and Guardrails lists - A yellow warning icon indicates a model approaching retirement. A red error icon indicates a retired, unavailable, or unsupported model. Point to the icon to view details.
- Flow builder - Warning or error icons appear on affected AI agent and Custom prompt check guardrail steps. Point to an icon to view the model status and recommended replacement.
- Home - A warning or error icon appears on an affected integration tile or next to its name in list view. Point to or select the icon to open the Model attention panel.
- Flows tab - Warning or error icons appear next to affected flows. Select an icon to open the Model attention panel for that flow.
AI agents and guardrails that aren't used in any flow appear only in their respective lists.
Review models that need attention
The Model attention panel lists models that need attention across the flows in an integration.
- Go to Home.
- Point to or select the warning or error icon for the integration. The Model attention panel opens.
- Review the following details:
- Model - The AI model used by an AI agent or guardrail.
- Status - The model's current lifecycle status.
- Recommended - The suggested replacement model, when available.
- Date - The model's retirement date, when applicable.
- Optional: Use the Status filter to view models with specific statuses.
What happens when a model is no longer supported
AI agents and guardrails using models marked Retired, Unavailable, or Not supported can't run until you select a supported model.
- Saving changes - You can't save changes to an affected AI agent or guardrail, including unrelated changes, until you select a supported model.
-
Testing and execution - Preview tests and flow runs are blocked before the model is called. Each affected record appears in the flow step's Errors list with one of these errors:
Run blocked: model "model ID" is retired on "date". Please choose a supported model and test your agent configuration.Run blocked: model "model ID" is DISABLED. Please choose a supported model and test your agent configuration.Run blocked: model "model ID" is not supported. Please choose a supported model and test your agent configuration.
- Cloning and deployment - Cloning integrations, installing templates, or pulling changes between environments fails if they include AI agents or guardrails using these models. Switch the model in the source, and then try again.
Models marked Migration recommended or Retiring on date continue to run normally until retirement. These warnings don't prevent you from saving changes or running flows.
To resolve execution failures, switch to a supported model, save the AI agent or guardrail, and retry the failed records from the flow step's Errors list.
Switch to a supported AI model
Switch an AI agent or guardrail to a supported model when its current model is approaching retirement or is no longer available.
Note: Only guardrails configured with Custom prompt check use an AI model.
Important: Switching an AI agent to another model resets its advanced settings to the new model's defaults. Settings that the new model doesn't support are removed. Review the settings before saving.
- Go to AI studio > AI agents or AI studio > Guardrails, and open the affected AI agent or guardrail.
- In AI model selection, select a supported model. If a recommended replacement is available, it appears below the field or in the Model attention panel.
- For an AI agent, select Continue in the Switch model dialog.
- For an AI agent, select the settings icon next to AI model selection to review AI model advanced settings.
- Test the updated configuration:
- AI agent - Select Test in the Preview pane.
- Guardrail - Select Send in the preview.
- Select Save & close.
Use the Celigo API
The same rules apply when you create or update AI agents and guardrails through the API. A request that saves an AI agent or guardrail with a retired, unavailable, or unsupported model fails with HTTP status 422 and the error code ai_model_retired, ai_model_disabled, ai_model_not_in_catalog, or ai_model_not_available. The error message names the model and, when one is available, the suggested replacement. Runs started through the API are blocked with the same Run blocked messages.