What is an AI Sandbox?

An AI sandbox is a private, disposable cloud workspace where AI agents run code. Instead of executing on your laptop, the agent works in an isolated environment with its own file system — so the experiment stays clean, your machine stays untouched, and every project gets its own space.

In plain English

Imagine hiring a contractor to renovate your kitchen. You wouldn't let them demo the walls with your laptop sitting on the counter. You'd give them a separate workspace — a garage, a workshop — where they can make a mess, test things, and only hand you the finished result.

An AI sandbox is that workshop. The agent builds your app in a clean cloud environment, runs the code, checks it works, and shows you a preview. Your laptop never hosts a half-finished experiment, and the agent never touches your other files.

The name comes from the kid's sandbox: a contained place where building and knocking things down is expected.

Why sandboxes exist

There are two reasons AI agents work in sandboxes, and both matter:

Safety first. An agent that installs packages and runs code is doing real things on a real system. If it ran on your laptop, a bad dependency or a broken script could touch your files. The sandbox keeps all of that on the other side of a glass wall.

Isolation second. When you build ten projects, each one needs its own clean room. A shared space lets project A's leftover files leak into project B. Orbit gives every project its own sandbox with its own file system — what happens in project A stays in project A.

What happens inside one

A single build runs a whole cycle inside the sandbox:

1. The agent installs dependencies and scaffolds the project.
2. It writes the app's source files.
3. It runs a typecheck and fixes whatever breaks.
4. It starts a dev server and probes it over HTTP.
5. You get a live preview URL — served from the sandbox.

The agent isn't claiming the app works; the sandbox proves it. Files are checked on disk, the server is probed over HTTP, and only then does the preview unlock. That verification loop is what separates a working build from a confident one.

How Orbit's sandboxes work

Orbit runs builds in cloud sandboxes with a few extra guarantees:

• One-hour sessions — enough for a full build cycle, with pause and resume so your work persists.
• Per-project locks — one writer at a time, so concurrent runs can't collide.
• Schema-isolated databases — each app's Supabase schema is separated, so multi-tenant data never mixes.
• Security scans — leaked secrets block the final ship instead of shipping a key out the door.

Your files and previews are durable even though the sandbox itself is temporary. Close the session, come back later, and the project is still there — the workshop gets cleaned, but the furniture you built stays yours.

Why this makes AI builders trustworthy

A tool that generates code without running it is only telling you a story. A tool that generates code in a sandbox, runs it, and verifies it is showing you receipts.

That's the real value of the sandbox: it turns "trust me, it works" into "watch it run." When you can click through a live preview that was typechecked and probed over HTTP minutes earlier, you're not betting on the model's confidence — you're looking at the actual output.

Try it and compare

Ask your current tool where its code runs. If the answer is vague, or the "preview" is a static screenshot, you have no sandbox — you have a mockup generator. Then ask Orbit to build something with a login and a database, and watch the preview come from a real running sandbox. That difference is the entire point.

FAQ

Q: What is an AI sandbox? A: An AI sandbox is an isolated cloud environment where AI agents run code. It keeps the agent's work away from your machine and your other files, and gives every project its own clean space.

Q: Why do AI app builders need a sandbox? A: Because AI output has to be verified, not trusted. A sandbox lets the agent install dependencies, run the code, typecheck it, and start a preview without touching your laptop or colliding with other projects.

Q: Is my code safe in an AI sandbox? A: Yes — isolation is the point. Each project gets its own sandbox with its own file system, and Orbit adds per-project locks and security scans so work doesn't bleed between projects or leak secrets.

Q: Can I keep using my app after the sandbox closes? A: Yes. Orbit keeps your files and preview durable, and the sandbox session can pause and resume. Your work persists even though the runtime environment is temporary.

Build it on Orbit

Orbit runs every build in an isolated cloud sandbox — typechecked, verified, and previewed before you see it, with your files kept safe and durable. Start building on Orbit.