Does OpenSandbox support syncing sandbox directories to S3? Any plans for built-in support?

[qiankunli]

Hi OpenSandbox team,

I'm currently evaluating OpenSandbox for building an AI agent sandbox environment.

In our use case, agents running inside a sandbox may generate or modify files during execution (for example under a directory like /workspace or /conv). Since sandboxes are typically ephemeral, we would like to persist these files to external object storage such as S3.

Our goal is to automatically synchronize a specific directory inside the sandbox to S3 so that files produced by the agent can be preserved even after the sandbox is terminated.

Example scenario:

• An agent runs inside a sandbox
• The agent reads/writes files under a working directory (e.g. /workspace)
• Files generated during execution should be synchronized to S3
• When the sandbox expires or is destroyed, the files are already persisted externally

I would like to ask:

1. Does OpenSandbox currently provide any built-in mechanism to synchronize sandbox directories to object storage such as S3?
2. If not, are there any recommended patterns for implementing this (for example using volumes, external sync services, or runtime extensions)?
3. Are there any plans on the roadmap to support built-in storage synchronization or persistent filesystem features for sandbox workloads?

Any guidance or best practices would be greatly appreciated.

Thanks!

[Spground]

Yes, OpenSandbox provides Volume to handle your use cases, see also https://github.com/alibaba/OpenSandbox/blob/main/oseps/0003-volume-and-volumebinding-support.md for more information. We are working about it .

[hittyt]

Hi @qiankunli ,

Thanks for your interest in OpenSandbox and for reaching out! This is a great question.

We absolutely agree that decoupling the sandbox lifecycle from the storage lifecycle is critical for Agentic workflows, and we are actively working on implementing these features.

To answer your specific questions:

1. Built-in mechanism for Object Storage (S3/OSS)
   Yes, we currently provide built-in support for mounting object storage via ossfs (which is compatible with S3-like architectures) in our Docker mode. You can mount it directly to your target working directory (e.g., /workspace). Support for the Kubernetes (K8s) mode is currently on the way.

2. Recommended Patterns
   While direct object storage mounting works, leveraging Kubernetes native Persistent Volume Claims (PVCs) might be a more robust and cloud-native pattern for your specific use case. By using a PVC backed by an S3 CSI driver (like s3-csi), you can handle the synchronization automatically at the infrastructure level, keeping the sandbox entirely ephemeral.

3. Roadmap for Volume Support
   We have already implemented PVC-type volume support for our Docker mode. Native K8s PVC support is on our near-term roadmap and is coming very soon. Once released, this will allow you to seamlessly bind any standard Kubernetes storage class directly to the sandbox instances.

Feel free to track our upcoming releases, and let us know if you have more details to share about your Agent orchestration setup. We'd love to hear more about your use case!