Your First AI Employee:
From Laptop to Cloud in 10 Minutes

What you're about to do

You're going to create a cloud computer for your AI agent. Not a Docker container on your laptop. Not a shared VM. A dedicated Linux machine in the cloud that the agent owns. It has its own filesystem, its own CPU, its own network. It runs when your laptop is closed. It costs $0.02/hour. And it takes 10 minutes to set up.

After this guide, your agent will be running on its own machine. You'll be able to assign it a task, close your laptop, and check the results tomorrow. That shift — from "AI tool I hold" to "AI employee that works" — starts here.

Three paths, same destination

The sandbox works with any agent. Pick the one you already use:

• Claude Code — SSH in, run claude, assign work in natural language. The most popular path. If you've used Claude Code on your laptop, this is identical except it runs on a cloud machine.
• Codex CLI — SSH in, run codex. Same pattern, different model. If you're in the OpenAI ecosystem, this is your path.
• API-driven — no SSH needed. Use the execute API to run commands remotely. For custom agents, LangChain scripts, CrewAI workflows, or any automation that calls the sandbox programmatically.

All three paths start the same way: create the sandbox. The sandbox is the computer. What you install on it is up to you.

The moment it clicks

Here's what happens when you finish this guide:

Your agent is running on a cloud machine. You SSH'd in, installed Claude Code (or Codex, or your custom agent), and told it to clone a repo and run the tests. It's doing that now. On the cloud. Not on your laptop.

You close your terminal. Nothing happens. The agent keeps working. You make coffee. The agent keeps working. You go to lunch. The agent finishes, commits its changes, and the results are waiting in /workspace when you check back.

That's the moment it clicks: this isn't a tool you use. It's an employee that works. It doesn't need your attention. It doesn't need your laptop. It just needs a computer. You just gave it one.

What's different from your laptop

You might wonder: "Why not just run the agent on my laptop?" You can. You have been. But the cloud machine changes five things:

1. It runs while you sleep. Close your laptop, the agent dies. A cloud sandbox keeps running. Kick off an overnight refactoring job at 11pm. Read the results at 8am.
2. You can have 10 of them. Your laptop is one machine. Ten sandboxes are ten machines. Ten agents working in parallel, each on their own task.
3. It's disposable. If the agent installs something bad or breaks the filesystem, you delete the sandbox and create a new one in 30 seconds. Your laptop has your real files.
4. You know what it costs. Every second is tracked. You know this task cost $0.08 in compute. You can budget, allocate, and report.
5. It has three interfaces. Terminal + browser + desktop. Your laptop gives the agent a terminal. The platform gives it a browser (Chrome with CDP) and a desktop (XFCE with VNC) too.

What to try first

Now that you have a sandbox, here are four tasks that demonstrate what's different about a dedicated machine:

1. Overnight refactoring (30 seconds to assign)

SSH in, start tmux, run Claude Code: "Refactor the auth module from callbacks to async/await. Run tests after each change. Keep going until all tests pass." Detach tmux (Ctrl+B, D). Close your laptop. Check results in the morning.

2. Dependency audit in isolation (2 minutes)

Clone an unfamiliar open-source repo on the sandbox. Install its dependencies. If something malicious runs during npm install, the blast radius is a disposable VM, not your laptop.

3. Parallel test matrix (5 minutes)

Create 3 sandboxes. Each runs your test suite against a different Python version (3.10, 3.11, 3.12). All three run simultaneously. Total cost: $0.02.

4. Browser research (5 minutes)

Create a Chrome browser container alongside your sandbox. The agent uses the browser for web research and the sandbox for processing. Two containers, two interfaces, working together.

What it costs

Less than a cup of coffee for an overnight run. Less than a penny for this quickstart. The economics are not a barrier. The only barrier was the 10 minutes you just invested.

What's next