To achieve this, you need co-design of the model and the harness. The first version ships with a post-trained model on top of Qwen, but we intend to support Nemotron 3.5 Lightning and more models in the future. Post-training also allows the orchestrator to escalate to a cloud frontier with user permission.