What if you could multiplex roughly 250 stateful agent sessions across eight Kubernetes worker Pods, then reactivate any one without losing its in-memory or filesystem state?The repository's demo reports 30x+ actor-to-worker oversubscription for that sample workload, with sub-second activation. It is a demonstration, not a production capacity guarantee.