A platform that takes your app from a prompt to a live URL with nothing to configure. It runs microVMs. A sandbox is a small computer, and a production service is the same one with push to deploy, previews and rollback.

Is a Pilots sandbox only for AI agents?

No. It is a whole Linux computer with root access, so it is for whatever a computer is for. Agents are one use, and so are test fleets, builds and code you do not trust.

Does Pilots use containers?

No. Every computer is a microVM with a kernel of its own, whether it is a sandbox or a production service.

Can a sandbox become a production service?

Yes. One command promotes a sandbox running your image into a service. Nothing is rebuilt, and it gains push to deploy, previews and rollback.

Does a computer keep running when nobody is using it?

No. It is suspended with its memory, uses no compute while it sleeps, and the next request wakes it with every process where it was.

Do I have to configure anything to deploy?

No. Your agent works out what the app needs, the database and background jobs included, and Pilots deploys all of it.

Do I need a Dockerfile to deploy?

No. Pilots recognises the framework in a directory and builds it. It uses a Dockerfile or a compose file when you have one.

Can an AI agent operate Pilots by itself?

Yes. Every operation is a command in the CLI, and the MCP server offers them to an agent directly, so nothing needs a browser.

What is Pilots?

Pilots is a platform that runs microVMs. You ask your agent, and your app goes from a prompt to a live URL with nothing for you to configure. A sandbox is a small computer that can do anything a computer can. A production service is the same computer with more on top, such as push to deploy, preview environments and rollback.

Install the CLI Why Pilots

One thing, used two ways

There is one primitive here, a computer. A sandbox is a computer you treat as disposable. A service is one or more computers behind a permanent address. Promoting turns the first into the second.

one computer, from scratch pad to production

$ pilot machine create my-shopNAME     STATE    CREATED   URLmy-shop  running  just now  https://my-shop.pilotrun.app$ pilot exec my-shop -- npm testPASS  test/checkout.test.js$ pilot promote my-shopSERVICE  URL                           IDmy-shop  https://my-shop.pilotrun.app  svc_7c1e94a0same address, now a production service

A computer: What the commands call a machine. A microVM with its own kernel, its own network, a disk, and an address made from its name. It runs a stock image with a shell, Node and Python, or an image built from your code.
A sandbox: A computer you can throw away. Somewhere to run a command, try a build, test a product or let an agent work, without any of it touching your laptop.
A service: One or more computers behind a URL, with releases, a health gate, replicas and rollback. A deploy replaces the computers and leaves the address where it was.

A computer that is there when you ask

Nothing boots. A new computer is restored from a snapshot of one that was already running, and a sleeping one is woken with its memory. That is why these are counted in milliseconds.

468ms create, median on metal

302ms wake, median on metal

403ms checkpoint resume gap, median on metal

These are medians from the production fleet, printed by the test battery on one of its hosts. Every number on this site carries its source. Hover one to see it.

What a computer does

Runs anything a computer runs

A computer is Linux with a kernel of its own and root access. Install a browser, a compiler or a database, and run code you have not read.

Sleeps when it is idle

A computer nobody is using is suspended with its memory and uses no compute. The next request wakes it, and every process is where it was.

Grows into production

Promote a sandbox and it gains push to deploy, preview environments, rollback and a domain of your own. Its address stays where it was.

Remembers a moment exactly

A checkpoint captures the whole computer, memory included. A restore puts that moment back, and a fork starts a new computer from it.

Deploys a directory

One command takes the directory you are standing in to a URL, behind a health gate, so a broken release never replaces a working one.

Answers to an agent

Every operation is in the CLI and offered over MCP, so an agent can create, deploy, checkpoint and promote without a person in a browser.

Questions people ask

Short answers. The pages behind them are the documentation.

What is Pilots?: A platform that takes your app from a prompt to a live URL with nothing to configure. It runs microVMs. A sandbox is a small computer, and a production service is the same one with push to deploy, previews and rollback.
Is a Pilots sandbox only for AI agents?: No. It is a whole Linux computer with root access, so it is for whatever a computer is for. Agents are one use, and so are test fleets, builds and code you do not trust.
Does Pilots use containers?: No. Every computer is a microVM with a kernel of its own, whether it is a sandbox or a production service.
Can a sandbox become a production service?: Yes. One command promotes a sandbox running your image into a service. Nothing is rebuilt, and it gains push to deploy, previews and rollback.
Does a computer keep running when nobody is using it?: No. It is suspended with its memory, uses no compute while it sleeps, and the next request wakes it with every process where it was.
Do I have to configure anything to deploy?: No. Your agent works out what the app needs, the database and background jobs included, and Pilots deploys all of it.
Do I need a Dockerfile to deploy?: No. Pilots recognises the framework in a directory and builds it. It uses a Dockerfile or a compose file when you have one.
Can an AI agent operate Pilots by itself?: Yes. Every operation is a command in the CLI, and the MCP server offers them to an agent directly, so nothing needs a browser.

Read why Pilots is built this way, or see it beside Railway, Render, Vercel, Modal, Fly.io and E2B.