Codex cloud, or Codex on a computer that stays on? Where to run it 24/7

What is Codex cloud, and what changed on 29 September?

Codex cloud runs coding tasks on computers OpenAI manages. Since OpenAI's DevDay on 29 September 2026 (TechCrunch's report), you set up an environment once (the repositories, dependencies, tools and access settings a task needs), publish it, and every new task starts from it in its own isolated workspace. OpenAI's cloud environments docs say you create and publish on desktop or web, then use it from web, mobile or the desktop app.

Every OpenAI fact on this page comes from those docs, OpenAI's pricing page, authentication, remote connections, non-interactive mode, agent approvals and its changelog. Other OpenAI docs are linked where they come up. There's also an older flow, Codex Cloud (Legacy). Code Review, Security Review and the existing Linear and GitHub integrations still run on it for now, and OpenAI says it plans to retire it.

A quick word on names. OpenAI's docs say Codex Cloud. Its pricing table says Codex cloud. The plan cards on the same page say Codex on the web, and plenty of people just search "codex web". It's all one product (we couldn't find any OpenAI page announcing a rename), so we call it Codex cloud here.

Will Codex keep working when I close my laptop?

Yes, with Codex cloud. The work runs on OpenAI's computers, and OpenAI's cloud environments docs say tasks can continue while your computer is asleep. The Codex CLI on your laptop won't, because it needs the laptop awake. A computer that stays on, with the CLI inside tmux, keeps going too.

On Hivra, what survives a closed laptop depends on where you start the run. On a paid Hivra plan the computer stays on and keeps your files, sessions and login. A Claude Code or Codex run you start inside tmux in the computer's Terminal tab keeps going after you close the laptop. On Claude Code, work you send through Telegram, connected in the agent's Telegram tab, runs on the computer, not in your browser.

Codex Remote is a fifth option with a catch. It lets you reach a Codex session on your own computer from your phone or another computer, but OpenAI's remote connections docs say the host has to run the ChatGPT desktop app on macOS or Windows, awake, online and signed in. Remote access stops if that computer sleeps or the app closes, and a Mac laptop with the lid closed needs an external display. For longer work OpenAI suggests a dedicated always-on Mac or Windows PC. The Codex agent on Hivra is the CLI, and we haven't tested Codex Remote on a Hivra computer.

Codex cloud, Codex on a Hivra computer, your laptop or a VPS: which should I use?

Pick by the job. One self-contained change to a GitHub repo goes to Codex cloud. Files outside GitHub and a shell you can rejoin go to a computer that stays on.

When is OpenAI's Codex cloud enough?

If the job is one self-contained change to a GitHub repo and your ChatGPT plan includes Codex cloud, use it. No server to look after. For that job Hivra can't beat it, and we won't pretend otherwise.

OpenAI's pricing page lists Codex cloud for Plus ($20 a month as of October 2026), Pro (from $100 a month), Business and Enterprise or Edu. An API key doesn't get you in. The same page says API key access has no cloud features at all. The Free and Go cards only mention Codex in the desktop app. One oddity: OpenAI's GitLab docs say the GitLab beta, which runs in Codex cloud, is open on every ChatGPT plan. So check that your own account shows Codex cloud before you count on it. For the dated Plus and Pro cards, see Codex pricing by ChatGPT plan.

Hardware isn't a reason to switch either. By default OpenAI gives each cloud task a VM with 2 vCPUs, 8 GiB of memory and 8 GiB of disk on Plus, and 4 vCPUs, 16 GiB and 32 GiB on Pro, Business and Enterprise. A Plus task VM has more memory than Hivra's $9.99 a month plan, which is 2 vCPU and 4 GB of RAM.

What does Codex cloud leave out?

GitHub is the big one. Cloud environments check out GitHub repositories, and OpenAI's docs list GitLab and self-hosted GitHub Enterprise Server as not supported yet. There is a separate GitLab integration in beta since 19 August 2026 (GitLab docs), but it starts legacy cloud chats, not the new environments.

Computer use and browser use aren't supported in cloud environments yet either. Skills that live in your repo come along. Personal skills on your own machine don't.

Then there's memory. A task's saved VM state can be recovered for up to 7 days after you last start a turn or resume it. After that it's gone, and a new task never picks up another task's uncommitted changes. Treat git as the only thing that lasts.

Sign-in is ChatGPT only. An API key works for the CLI, the desktop app and the IDE extension, but not for cloud. If you log in with email and password, OpenAI wants MFA switched on first (authentication page).

Usage comes out of the same pool as everything else. OpenAI's pricing page says local messages and cloud chats share one allowance, cloud tasks can eat more of it, and weekly limits can kick in. You also pick, per environment, whether a task can reach the internet and which domains. How long can one cloud task run? None of the pages we read give a number, so we won't guess.

When does a computer that stays on make more sense?

When the work isn't a tidy task on a GitHub repo. Think files that aren't in git, or a shell you want to come back to. Codex cloud starts each new task in its own isolated workspace and, by default, recovers a task's saved state for up to 7 days. On a computer that stays on, your files and half-finished sessions are still there when you come back.

Codex cloud and the CLI both sign in with ChatGPT, and one cloud environment can hold several GitHub repositories. So neither of those settles it. Still choosing between agents? Claude Code vs Codex for 24/7 work compares them.

What Hivra adds over a plain VPS is mostly setup you skip. The dashboard starts Codex's device-code sign-in for you. A Sign in with ChatGPT window shows a code, you type it into OpenAI's sign-in page with your own account, and Codex on the computer is signed in. No terminal. OpenAI's rule still applies, so switch on device code login (marked beta) in your ChatGPT security settings first. tmux is already installed, and the Terminal tab under Computer is a plain shell. Paid plans aren't paused for inactivity. You reach the computer from a browser, on a phone too. There's no native app.

Now the trade-offs, because there are real ones. Hivra is a second bill on top of your ChatGPT plan, and it adds no Codex allowance. Codex on a Hivra computer still bills through your own ChatGPT or OpenAI account, from the same shared pool.

You also get less hardware per dollar. Hostinger's KVM 2 is 2 vCPU and 8 GB of RAM for $8.99 a month on its 24-month plan, paid upfront. Hivra's $9.99 plan is 2 vCPU and 4 GB of RAM. And Agent 37, which also hosts Codex, works out to $4.76 a month for an always-on 2 vCPU, 4 GB shape on its metered Cloud API, as long as you keep a prepaid balance topped up (Hivra vs Agent 37 has the figures).

Two smaller ones. Hivra installs the Codex CLI version it has tested and turns off the startup update check, so you can trail OpenAI's newest release by a bit. And there's no SSH into the underlying server. You reach the computer from a browser, which also means you can't point the ChatGPT desktop app's SSH-host feature at it (remote connections docs). A VPS can do that.

If you run Codex through Hivra's chat, it runs with approvals and the sandbox bypassed by default on the computer. That bypass is Codex's --dangerously-bypass-approvals-and-sandbox flag. OpenAI's approvals docs say to use it only inside a controlled environment, and the skip-permissions guide sets it beside Claude Code's. The Permissions setting on the agent's Manage tab narrows that to Limited (workspace sandbox) or Read-only.

How do I run Codex 24/7 on my own server or a Hivra computer?

Install the CLI, sign in, and start it inside tmux on a machine that stays awake. The Codex CLI is an interactive terminal agent that runs as a child of your shell, so a laptop that sleeps, an SSH connection that drops or a terminal you close takes the run with it. tmux protects against disconnects. It can't keep a suspended laptop computing. Not sure your setup has a gap? Run it through the agent survival check.

On a VPS or a home server. One example: Hostinger's VPS pricing page lists KVM 2 with 2 vCPU, 8 GB of RAM and 100 GB of NVMe at $8.99 a month on its 24-month plan, paid upfront (about $216 in one payment by our arithmetic), renewing at $14.99 a month, as of October 2026. Hivra vs Hostinger has more. Then:

OpenAI's authentication page marks device-code sign-in as beta, and you have to switch on device code login in your ChatGPT security settings first (or in workspace permissions, if an admin runs your workspace). A rented server is yours to look after, too: Node and CLI updates, OS patches, disk space, SSH keys and the firewall.

For a one-shot job, codex exec runs a prompt and exits when it's done. Add nohup and the run survives your logout:

By default codex exec runs in a read-only sandbox, so it needs --sandbox workspace-write to edit files (OpenAI's non-interactive mode docs). That sandbox keeps network access off by default (approvals docs), so the run can't push or open a PR. It suits fire-and-forget jobs. To steer mid-task, use tmux.

On a Hivra computer you skip the install. Launch Codex from the Codex agent page, sign in with ChatGPT in the window the dashboard opens, then open the Terminal tab, under Computer, and start Codex inside tmux. The tmux cheat sheet lists the keys. To pick a chat back up after Codex exits, see how to resume a Codex session. Hivra runs the official Codex CLI on a private virtual machine. Plans are on the pricing page, with a 7-day money-back guarantee on card payments. Hivra is independent and is not affiliated with OpenAI.