Close the Tab. It Keeps Working.

Here's something every developer who uses AI coding agents has experienced:

You give the agent a task — a refactor, a migration, a complex multi-file change. It's going to take 20 minutes, maybe an hour. You start it running and then... you wait. Your laptop is committed. You can't close it, because if the terminal dies, the agent dies. You can't start another heavy task, because the agent is already using your CPU. You're babysitting a process.

Or worse: you want to run an agent overnight. A big refactor. A full test suite across 10 branches. But your laptop needs to stay open, plugged in, and connected to WiFi all night. If it sleeps, the agent stops. If you're traveling, you can't. If your partner closes the laptop thinking you left it open by accident — game over.

This is the fundamental limitation of running AI agents on your own machine: your laptop is the runtime, and your laptop has a life of its own.

The API/SDK Platforms Don't Solve This Either

You might think cloud sandbox platforms like E2B or Modal solve this. They don't — at least not for you, the person who just wants to run their agent.

E2B and Modal are APIs. They're built for developers who are building products that need sandboxed code execution. You write code that calls their SDK, creates a sandbox, runs code inside it, and retrieves results. That's great if you're building an AI platform. It's useless if you just want to boot Claude Code and let it work on your repo.

There's no "pick an agent, get a terminal, close the tab" experience. You'd have to write the integration yourself.

What Walk-Away Compute Actually Looks Like

With zipbox.ai, the experience is:

1. Open the page. Browser-native. No install, no CLI, no SDK.
2. Pick your agent. Claude Code, Codex, Cursor, Grok — any of 9 supported agents.
3. Get a terminal in seconds. A pre-warmed Firecracker microVM boots instantly with the agent already installed and running.
4. Give it your task. Full root. Auto-approve. Let it sudo, rm -rf, install kernel modules — whatever it needs.
5. Close the tab.

That's the key moment. Step 5. Close the tab.

The agent keeps working. It's running on a cloud machine, not your laptop. Your laptop is free — close it, put it in your bag, go to sleep, go to a meeting. The box is in the cloud and it doesn't care what your laptop does.

6. Come back when you're ready. Open zipbox.ai, find your box, see the results.
7. Pay for the seconds it ran. A 10-minute task: about 3¢. An overnight build: less than a dollar.
8. Nuke the box. Everything wiped. Nothing logged. Gone.

Why This Is Different From Everything Else

The difference isn't just "cloud vs local." It's the combination of cloud + zero-setup + disposable + metered. Codespaces is cloud, but it's a persistent IDE — heavy, expensive, designed to be a long-lived environment. zipbox is a disposable sandbox — you boot it, use it, nuke it. Different tool, different job, different price (cents vs dollars).

The Use Cases This Unlocks

Walk-away compute isn't a feature. It's a behavioral shift. Here's what becomes possible:

• Overnight refactors. Kick off a major refactor at 11pm, close your laptop, go to sleep. Wake up to a finished branch. Cost: ~60¢.
• Parallel experiments. 10 branches, 10 agents, 10 sandboxes — all running simultaneously in the cloud. Your laptop is free for actual work. Cost for an hour of 10 agents: ~$1.
• "Let me just try this." Want to see if an agent can handle a task? Spin up a box, give it 5 minutes, see what happens. If it fails, nuke it. Total cost: 1 cent. No commitment, no cleanup, no risk.
• Travel-friendly. On a plane with spotty WiFi? Start a task on zipbox from your phone, close the browser, land two hours later, check results. The box ran the whole time.
• Team handoffs. Start a task, share the box URL with a teammate, they pick up where you left off. No "is it done yet?" — just check the box.

The 30-Second Test

If you're reading this, try it right now:

1. Go to zipbox.ai
2. Pick an agent
3. Give it a small task
4. Close the tab
5. Come back in a few minutes

The agent kept working. Your laptop didn't. That's the point.