Skip to main content

Model Lifecycle Management

LLM providers regularly deprecate and retire models. MeetLoyd's Model Lifecycle Manager ensures your agents stay operational by providing advance warnings, automatic upgrade paths, and full control over how your organization responds to model changes.

How It Works

Every LLM model in MeetLoyd has a lifecycle state:

StateMeaning
AvailableModel is fully operational
PreviewNew model available for early testing
DeprecatedStill works, but a better successor exists
RetiringWill stop working on a specific date
RetiredNo longer available -- API calls will fail

When a model's state changes, MeetLoyd can automatically notify you, upgrade your agents, or pause them -- depending on your settings.

Lifecycle Indicators

In the Model Selector dropdown (when creating or editing agents), you'll see visual indicators:

  • Amber "Deprecated" badge -- model still works but you should plan to upgrade
  • Red "Retiring" badge with date -- model will stop working soon
  • Red "Retired" badge -- model is disabled and cannot be selected
  • Blue "Preview" badge -- new model available for testing

If you have an agent using a deprecated or retiring model, a warning banner appears with upgrade recommendations.

Notifications

EventPriorityChannels
Model deprecatedNormalIn-app, Email
Model retiring soonHighIn-app, Email, Slack
Model retiredUrgentIn-app, Email, Slack
Agents auto-upgradedNormalIn-app, Email

Each notification includes which model is affected, how many of your agents use it, the recommended successor, and a direct link to update.

Auto-Upgrade

When auto-upgrade is enabled and a successor model is defined, MeetLoyd updates the agent's model configuration automatically. The agent continues running on the new model at its next execution with no downtime.

When to Use Auto-Upgrade

Auto-upgrade works best for standard workloads. If you have agents with model-specific prompt tuning, consider using Notify & Warn instead and upgrading manually after testing.

Execution Guard

As an additional safety net, MeetLoyd checks the model lifecycle state before every agent execution:

  • Retired models are blocked -- the execution returns an error with the recommended successor
  • Deprecated models log a warning but continue executing normally

This ensures agents never silently fail due to a retired model.