The system runs both the agent harness and the underlying AI model locally to process tasks without needing external servers. If a specific task requires more advanced reasoning or research beyond the on-device capacity, the software requests user permission to tap into a cloud model for assistance.
