Cowork Shifts AI Model Inference to Cloud for Better Performance

The Cowork platform has transitioned from running model inference and tool calls in a local VM to executing these processes in the cloud. The previous approach, while secure, imposed significant disk, battery, and performance costs on users’ devices. The new cloud-based model addresses these issues by providing dedicated sandboxes for each session, eliminating the need for local VM execution. This change also allows users to maintain active sessions even when their devices are closed, such as when using Cowork from a phone. The update aims to enhance performance and usability while maintaining security and capability.