Skip to main content

00Managed agent sandbox cloud

Persistent sandboxes for agents that keep going.

Run without a platform-imposed maximum. Pause when idle, then restore the same encrypted rootfs on fresh compute.

ttl=0 supportedencrypted durable stateApache-2.0 runtime
Sandbox state maponline

Replaceable compute

alloc/sb-agent-7f29gVisor

Durable sandbox identity

Identity
sb_agent_01
Rootfs
checkpointed
Snapshot
rootfs
Objects
encrypted

Fresh compute

alloc/sb-agent-a914same identity

LIFETIME

No fixed maximum

ROOTFS

Snapshot + restore + fork

DURABLE LAYER

S3-compatible storage

CLOUD COMPUTE

$0.02 / GiB-hour

01One identity, many runtimes

A sandbox is more than an allocation.

Compute is replaceable. Sandbox identity and storage are not. Sandbox0 separates them so an agent can stop paying for idle compute without throwing away the workspace it built.

01

RUN

Let the agent finish the job.

No platform-imposed maximum runtime. Set ttl=0 and hard_ttl=0, or choose your own pause and deletion boundaries.

+ compute attached

02

PAUSE

Release compute, keep the sandbox.

Checkpoint the encrypted writable rootfs before releasing the runtime allocation.

+ state durable

03

RESUME

Continue on fresh compute.

Restore the same sandbox identity and rootfs into a fresh allocation when the agent is ready to work again.

+ identity preserved

03Agent-native workflow

The same controls for humans and agents.

Codex, Claude Code, Cursor, or any shell-capable harness can use the s0 CLI from registration through observability. No dedicated plugin is required.

  1. 01

    Register & authenticate

    Device login, teams, and scoped API keys.

    $ s0 auth login
  2. 02

    Build & operate

    Sandboxes, commands, files, services, templates, and snapshots.

    $ s0 sandbox exec <id> -- npm test
  3. 03

    Observe

    Stream logs and events; query bounded runtime metrics.

    $ s0 sandbox logs <id> --watch
Cloud quickstartPython 3.9+
$ s0 auth login
$ export SANDBOX0_TOKEN="$(s0 apikey create --name agent --role developer --raw)"
$ pip install sandbox0
import os
from sandbox0 import Client

client = Client(token=os.environ["SANDBOX0_TOKEN"])
with client.sandboxes.open("default") as sandbox:
    result = sandbox.run("python", "print('Hello from Sandbox0')")
    print(result.output_raw, end="")

Agent entry point

/llms.txt

Raw Markdown

/docs/sandbox.md

Managed or self-hosted on Nomad

Keep the workload model portable.

The same APIs, SDKs, templates, and s0 CLI target Cloud or the Apache-2.0 runtime on your dedicated Nomad clients.

Self-hosting Docs

04Ready when your agent is

Give every agent a workspace that survives the run.

Start on Sandbox0 Cloud and keep the state your agent needs without operating the runtime platform yourself.