froog / Use cases

Run Claude Code in the cloud

Give Claude Code its own Linux machine and let it work unattended: install packages, run Docker, open Chrome, start servers. Nothing it does touches your laptop, and you can watch or take over at any time.

Why a cloud VM for a coding agent

  • Safe to let it run. The agent has root on a machine of its own, so it can install and change anything without asking, and the worst case is a VM you delete.
  • It keeps working when your laptop sleeps. The sandbox stays on until you stop it.
  • Everything it needs is installed. Claude Code, Codex, Gemini CLI, OpenCode, Amp, Cursor and pi, plus Node, Python, Go, Rust, Java, Docker and Chrome.
  • Run many at once. One sandbox per task, each with its own disk, ports and desktop.

Set it up

Install the CLI and start a sandbox (about a minute):

curl -fsSL https://froog.tech/install | sh
froog login
froog new task-1

Give Claude Code your key once, and every sandbox you create has it. Or SSH in and sign in with your Claude subscription; the login stays on the disk, so forks keep it.

froog env set ANTHROPIC_API_KEY
# or: froog ssh task-1, then run claude and log in

Put your code in and give it a task. Output streams to your terminal:

froog exec task-1 -- 'git clone https://github.com/you/app'
froog prompt task-1 "in ~/app, make the failing tests pass and commit"
froog prompt task-1 "add a /health endpoint" --agent codex   # or gemini, opencode, amp, cursor, pi

Watch it, check its work, and see the app it built:

froog desktop task-1        # its screen, in your browser
froog exec task-1 -- 'cd app && git log --oneline -5'
froog host task-1 3000       # an HTTPS URL for its dev server
froog stop task-1            # disk kept; froog start task-1 picks up where it left off

It can use the desktop too

Every agent in a sandbox can see and drive its Linux desktop through a built-in computer MCP server: screenshots, clicks, typing and Chrome navigation. Ask froog prompt task-1 "open example.com in Chrome and tell me the page title" and watch it in froog desktop.

What it costs

Running time is billed per second, only while the sandbox runs. The default size (4 vCPU, 8 GB) is $0.0274 an hour, so a 4-hour session costs about 11 cents, and a sandbox left on all month is $20. Model usage is billed by your model provider, not by froog. See pricing.

Questions

Can I use my Claude subscription instead of an API key?

Yes. SSH into the sandbox with froog ssh NAME, run claude and log in. The login stays on the sandbox's disk, so restarts and forks keep it.

Which coding agents are installed?

Claude Code, OpenAI Codex, Gemini CLI, OpenCode, Amp, Cursor CLI and pi, the newest release of each when the image was built.

Is it safe to let Claude Code run unattended?

The agent works on its own VM with its own kernel and disk, not on your computer, and has no route back to your machine unless you open one. Give it only the keys it needs.

7 days free: 25 hours of running time, 2 sandboxes at once. Then from $0.0137 an hour, billed per second.