Codex in the cloud

Codex in the cloud

A cloud VM per project with the Codex CLI signed in, your repo cloned and your services on preview links.

To run Codex in the cloud on a machine you control, use a VM per project. homemade starts a Linux VM with the Codex CLI installed and signed in to your ChatGPT plan or OpenAI API key, your repo cloned and your services running behind preview links. You can watch it from a browser or your phone, run several at once on different branches, and the VM sleeps when idle.

The facts

What every VM comes with.

FactDetails
CodexThe Codex CLI is installed on every VM. Use your ChatGPT plan or an OpenAI API key.
MachineA Linux VM with its own kernel and disk. Sizes from 0.5 vCPU / 1 GB up to 16 vCPU / 32 GB.
AgentsClaude Code, Codex, OpenCode, pi, Kimi and Grok installed. Sign in once and every new VM starts signed in.
LinksA URL for every port a service listens on. Public, team only or owner only.
SleepSleeps when idle and wakes where you left it, files and terminals included.
AccessBrowser terminal (phone too), SSH, VS Code, Cursor, the CLI, an MCP server and an HTTP API.
PriceHobby is free: 25 VM-hours a month, 1 VM at a time, nano or tiny. Pro is $20/mo with $20 of usage included, then from $0.012 an hour (nano) to $0.18 (large) by size. Sleeping VMs don't use hours.

How to start

  1. Sign in once

    Open Account → AI tools and sign in with your ChatGPT plan, or add an OpenAI API key.

  2. Start a VM per task

    Connect a GitHub repo (or start from a template). Each VM starts with the repo cloned at the branch you pick, your install commands run and your services started.

    Terminal
    homemade vm create --branch fix-auth --wait
  3. Run Codex

    Open the VM's terminal in the browser or over SSH and run codex. The app it works on is already running on its preview link.

    Terminal
    homemade ssh misty-fern-142codex

When something else fits better

homemade isn't the right tool for everything.

  • Tasks from ChatGPT only

    If you only start tasks from ChatGPT and want results back as pull requests, OpenAI's Codex cloud is built in and needs no setup.

  • One box, run by you

    If you want one long-lived server you look after yourself, a plain VPS with tmux works and costs less at full uptime.

Questions

Can I run the Codex CLI on a cloud VM with my ChatGPT plan?

Yes. Every homemade VM has the Codex CLI installed. Sign in once under Account → AI tools and every new VM starts signed in.

How is this different from Codex cloud?

Codex cloud runs tasks on machines OpenAI manages. On homemade the VM is yours: you can open a terminal next to the agent, run the app, open it on a preview link, install anything and keep the VM between tasks.

Can I use Codex and Claude Code on the same project?

Yes. Every VM has Claude Code, Codex, OpenCode, pi, Kimi and Grok installed, so you can try the same task with two agents on two VMs.

Can Codex start homemade VMs itself?

Yes. The homemade MCP server (homemade mcp) gives Codex tools to create VMs, run commands in them and read the results. Add it with codex mcp add homemade -- homemade mcp.

Read on

Try it on the free plan

Hobby is free. Connect a repo and your first VM starts from it.