AI model router for agents. Checks whether a local model can handle the task before calling cloud inference. LOCAL/CLOUD verdict saves cost on every call.