# Make a task’s execution boundary explicit

Canonical: https://savrn.com/cloud/sandboxes

SAVRN Cloud · Coming soon. Public documentation and a browser simulation are available; connected services are not yet available.

Research assistants and automated workflows need a clear boundary around the tasks they execute. SAVRN Cloud’s planned sandbox service would make inputs, resource limits and permitted behavior part of that boundary. The current preview runs only built-in browser tasks. Connected sandboxes and customer-code execution are Coming soon.


## Begin with a narrow task

The preview offers small operations such as inspecting text or CSV shape, validating JSON and summarizing input structure. Choose a task, provide public or invented input and inspect the recorded result. These examples help a team discuss task packaging without executing arbitrary code, starting a container, contacting a network or using a GPU.

## Record limits alongside results

A sample sandbox includes displayed duration and memory choices, together with controls to run or stop the local task and export its record. These are demonstration settings, not evidence of enforced infrastructure isolation. A connected release would need to define what each limit measures and what happens when execution exceeds it.

## Treat permissions as part of the workload

Future tasks may need files, package dependencies, credentials or outbound access. Each requirement expands the execution boundary and must be explicitly qualified. The intended design needs separate decisions for image provenance, secret delivery, network destinations and artifact retention. A trusted internal task is not sufficient evidence that untrusted customer code can run safely.

## Preserve a useful completion record

A researcher should be able to identify the task version, supplied inputs, exit condition and artifacts produced. An operator needs enough evidence to distinguish cancellation, a resource limit and an application failure. The preview illustrates a small result record; connected execution will need durable, access-controlled evidence and a recovery policy appropriate to the workload.

## What you can explore today

Built-in text, CSV and JSON tasks run locally in the browser with simulated sandbox records.

## Before this service launches

Connected sandboxing requires tested isolation, enforceable resource limits, controlled secrets and network access, and supported cleanup.

## The customer outcome

The intended outcome is a bounded task whose permissions, execution result and artifacts can be reviewed without guessing what happened inside it.

## Related documentation

- [Sandboxes overview](https://savrn.com/cloud/docs/sandboxes-overview)
- [First sandbox run](https://savrn.com/cloud/docs/sandboxes-first-run)
- [Images, secrets and egress](https://savrn.com/cloud/docs/sandboxes-images-secrets-egress)
- [Explore sandboxes](https://savrn.com/cloud/console/#/sandboxes): Run a built-in sample task and inspect its record.