Introducing Virtual Models on AI Gateway

Virtual Models give a task a stable model ID, such as **vmc/coding-agent**, while AI Gateway centrally controls its model, fallbacks, priority, and Zero Data Retention settings. Changes propagate to callers within minutes without code changes or redeploys.

For coding agents, assign IDs by task and use a candidate virtual model to run evals before changing production. Team roles control who can edit configurations, and callers cannot override centrally set routing or data-retention requirements.

Provider-specific options stop applying when the underlying provider changes, so review settings during a switch. The article recommends testing candidates but does not specify a standard evaluation method.