DOCSGetting started
DOCS

Getting started

Sandbox Router runs your code in sandboxes: isolated, persistent virtual machines that deploy like containers. Give a coding agent its own machine, or put an MCP server online behind an authenticated endpoint. You pay for what they use from a prepaid balance.

This guide takes you from a new account to your first running sandbox, all in the console.

Sign in

  1. Go to console.sandboxrouter.io and choose Sign up.

  2. Enter your email address and a password, and choose Create account. If the page offers Continue with GitHub or Continue with Google, you can use that instead.

  3. If you signed up with email, open the link in the verification email. It verifies your address and signs you in.

You start with a personal workspace, and you are its owner. Sandboxes, MCP servers, API keys and the balance all belong to a workspace. To create another workspace or invite people to one, open Settings.

Add funds

Sandbox Router is prepaid: usage draws down the workspace's balance, and sandboxes run only while the balance is above $0 and a card is on file.

  1. Open Usage & billing.

  2. In the Auto-recharge section, enter an Amount of credits from $5 to $2,000, and choose Add funds.

  3. The first time, Stripe Checkout opens to take the payment and save your card. Later top-ups charge the saved card.

To keep the balance from running out, turn on Auto-recharge in the same section: when the balance falls below the amount you set, your card is charged to bring it back up to your target.

A sandbox on Native bills for its vCPU, memory and disk while it runs. See pricing for the rates, and Billing and usage for how the balance works.

Create your first sandbox

  1. Open Sandboxes and choose New sandbox.

  2. Under 1. Choose a provider, keep Auto: Sandbox Router picks where the sandbox runs.

  3. Under 2. Choose an image, keep Default: Alpine Linux with bash, coreutils, Node and Docker. Leave Command empty: the sandbox then waits for the commands you run in it.

  4. Under 3. Configure, keep the Size at M (1/2 vCPU, 4 GiB of memory, 8 GB of disk). The estimate beside the form shows what the size costs. Optionally add a label such as name = first-sandbox: a name is a label.

  5. Under 4. Network, leave Allowed outbound hosts empty to let the sandbox reach every host, or list the hosts it may reach, one per line. Connectors shows how a sandbox gets your API keys and tools without ever holding the keys.

  6. Choose Create sandbox.

The create holds until the sandbox runs, and its page opens.

Use the sandbox

On the sandbox's page:

  • Run a command runs a command, such as node --version, with a time limit. It runs one executable with its arguments, without a shell, so quote arguments that contain spaces. Its output shows below it.

  • Runtime output shows the sandbox's log: each state it reached, and why.

  • Get URL shows a port's URL, its provider's own, where the provider gives one (see Port URLs). The Network card lists the hosts it may reach, and Edit changes them.

  • Take snapshot has its provider take a snapshot, where the provider keeps them. The Snapshots card lists them.

  • Stop stops the sandbox, and Start starts it again.

A running sandbox stops by itself after 15 minutes with no call in flight. While it is stopped, it bills for no vCPU, memory or disk, and it stays stopped until you start it.

To put your own files in it, use the CLI: sandboxrouter sandboxes files put <sandbox> /workspace/app.js app.js writes one file.

When you no longer need the sandbox, delete it from its row's menu in Sandboxes.

Next steps

  • The CLI: install sandboxrouter, connect it to your workspace, and run sandboxes from a terminal.

  • Sandboxes: sizes, commands, files, snapshots, URLs and the rest of a sandbox's life.

  • Routing and policies: cost and latency preferences, regions, provider accounts, creation failover and policy layers.

  • Images and the registry: push your own image to registry.caas.sh and run it.

  • MCP servers: put an MCP server online behind an authenticated endpoint.

  • Connectors: give your sandboxes tools, MCP servers and APIs through your workspace's account or each user's own, with keys added at egress.

  • Browser: give the AI tools in your sandboxes a web browser, signed in to websites as each of your users.

  • Logins: keep your users' sign-ins to websites, and let them sign in by link.

  • Billing and usage: the balance, auto-recharge, spend alerts and the usage report.

  • The API: automate all of this with an API key.