GCP agent workspaces: compute on demand, files that persist
When configured to leverage Filestore, GKE storage management happens behind the scenes: Every time GKE launches a sandbox for a new agent task, Filestore automatically allocates and attaches a dedicated, isolated file workspace to that environment in milliseconds.
GCP offers an architecture that decouples compute from storage:
Runtime (compute environment): Provided by Agent Substrate on GKE or GKE Agent Sandbox. They allocate isolated CPU and memory, providing the compute sandboxes agents need to safely execute dynamic code and install third-party packages.
Storage (workspace): Agent volumes provided by Google Cloud Filestore serve as the persistent filesystem.
These two layers work together automatically underneath. When GKE starts a compute sandbox (runtime) for a new agent task, the system automatically mounts a dedicated Filestore agent volume to it within milliseconds. Once the agent becomes idle, the system can proactively suspend the GKE sandbox to save expensive compute costs, while the agent’s working state and files remain safely stored in Filestore, until a new task arrives and compute resources resume instantly. This is quite useful for enterprises.