Google Cloud has announced Filestore agent volumes, a fully managed capability in Google Cloud Filestore that provides high-performance, elastic file storage for agentic workloads. The announcement targets platform teams running AI agents at scale, where storage has become a bottleneck. Today those teams typically stitch together custom workarounds: shuttling state between agent sandboxes and centralized storage, manually managing local disks, or operating self-hosted file systems. As fleets scale to thousands or millions of concurrent sessions, those approaches force trade-offs between cold-start latency, operational complexity, and the cost of idle, pre-allocated storage. The new capability is designed to meet the requirements of dynamic agent workloads: strict workspace isolation, instant session resumption, elastic pay-per-use economics, and fluid multi-agent collaboration.

The storage layer is co-designed with Agent Substrate and complements the dedicated compute environments provided by Agent Substrate on GKE and GKE Agent Sandbox. The core mechanism is dynamic provisioning: 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. Platform teams no longer need to manually create, attach, or tear down storage volumes for individual agent runs; the system handles the entire volume lifecycle automatically as the agent fleet scales up and down.

Google Cloud highlights four capabilities. Granular isolation and enterprise guardrails enforce strict boundary controls and per-workspace access permissions, keeping agents within their designated directories and isolating dynamic toolchains across tenants. Sub-second session resumption works because agent volumes attach and detach in milliseconds, letting orchestrators aggressively suspend idle sandboxes to save compute and resume instantly when new tasks or user inputs arrive. Smart lifecycle economics use pay-per-use pricing with automatic lifecycle tiering: when agents are not actively reading or modifying files, idle workspace state automatically shifts to lower-cost storage, avoiding the waste of pre-allocated fixed-size high-performance storage. Multi-agent collaboration is built on native Read-Write-Many (RWX) support with POSIX file locking, allowing a single shared workspace to be attached across multiple agents that safely co-author, test, and review files concurrently with file-level consistency and protection against write conflicts.

The post describes three production use cases. Software engineering and coding sandboxes can spin up thousands of isolated workspaces where agents install libraries, write multi-file patches, run build tools, and execute unit tests using standard POSIX file semantics with no storage-specific customization. Collaborative multi-agent swarms, such as a lead orchestrator delegating to research, code generation, and validation sub-agents, can directly share a unified file tree. Interactive long-horizon workflows, including user-in-the-loop applications that require asynchronous approval or run multi-hour data analysis pipelines, can suspend idle sandboxes and resume execution on demand with sub-second responsiveness.

Filestore agent volumes are now available to all Google Cloud customers for non-production workloads, with GA support for production workloads available via allowlist. The offering includes out-of-the-box integrations with Agent Substrate on GKE and GKE Agent Sandbox. Access is requested through a form, with further details in the Filestore and GKE documentation. The intended users are teams building Agent-as-a-Service platforms, scaling coding assistants, or deploying enterprise agent fleets.