Sandboxes

A sandbox is an isolated GPU container with a running SSH server. It lives until its timeout, its idle timeout, or until you stop it. The working directory is /workspace; volumes mount under /mnt.

Create

ts sandbox create --gpu a100-80 --image pytorch --name train-01 --lease 8 --idle 30 --volume datasets --secret huggingface

Or click New sandbox in the console. Flags: --gpu one of t4, l4, a10g, l40s, a100-40, a100-80, h100, h200, b200 · --count 1–8 GPUs · --image ubuntu, cuda, pytorch · --lease hours (max 24) · --idle minutes.

Connect

ts sandbox ssh m-81de342e82            # interactive shell through the gateway
ts sandbox exec m-81de342e82 -- nvidia-smi
ts sandbox stop m-81de342e82

SSH goes through the Tengsuan gateway, so no ports are exposed on your side and the endpoint closes when the sandbox stops. Password and key auth are both supported; pass --ssh-key ~/.ssh/id_ed25519.pub at creation.

Console shell

Every sandbox page has a shell that runs commands inside the container and streams the output, useful for quick checks without SSH.