Managed OpenAI Codex hosting: what is included, and what it costs

A month of one Codex agent: managed vs self-hosted

Model usage is left out because it costs the same on both paths. Hours follow the same self-hosting estimates as our OpenClaw cost guide.

Ways to run a Codex agent around the clock

What managed Codex hosting includes

Qoren gives a Codex agent a dedicated cloud environment and keeps it online. The environment has a persistent filesystem so work carries across runs, isolation so each agent is sandboxed from the rest, and secret storage so keys and tokens are never baked into a prompt or a repo. You configure the agent and its spend policy in the dashboard rather than over SSH.

• A persistent, isolated environment with a real filesystem.
• Runs through codex exec, one run per message or task, from chat or from webhook triggers.
• Scheduled tasks through Qoren's platform scheduler, which is still rolling out to accounts.
• Secret storage for model keys, source access, and tool credentials.
• Activity logs, backups, and repair from the console, with a public status page.
• A hard cap on managed model spend, or your own key or ChatGPT plan instead.

• New to Codex? What the Codex CLI is and how it runs

What it costs

Plans start at $39 a month, which hosts one agent all month; larger plans host more agents per environment. Each plan's monthly credits pay for that hosting and a share of managed model usage, and top-ups, which never expire, cover the rest. Model usage is billed by how much the agent reads and writes, and those tokens cost the same wherever Codex runs. Qoren does not change the model price; it gives you three ways to pay it.

• Managed key: usage draws on your plan's credits and stops at a hard cap you set.
• Your own OpenRouter key, on every plan: OpenRouter bills you directly, with no managed usage cap.
• ChatGPT sign-in: a Codex agent can run on your ChatGPT plan instead of a model key.

• Compare Qoren plans
• Bring your own OpenRouter key (BYOK)
• Sign a Codex agent in with ChatGPT

What makes hosting Codex different

Most agent runtimes are a model in a conversation loop. Codex is a runtime that executes: it drives the OpenAI Codex CLI to read files, run commands, and edit a working tree through codex exec. That changes what hosting has to provide. A chat agent needs a process and a key. A Codex agent needs a durable filesystem, the tools it shells out to, room to run tasks that take minutes, and isolation so its commands never touch anything you did not scope. Host it on a laptop and the working tree dies with the lid; host it on a bare VPS and you own every one of those concerns yourself.

Good Codex workloads for managed hosting

Codex earns its keep on recurring, code-shaped work that should happen whether or not anyone is watching. An agent can triage a repository, open or review pull requests, run a refactor or dependency bump, regenerate documentation from source, or turn a backlog item into a draft change. These are jobs with a trigger and a deliverable, which is the shape managed hosting is built for: the agent wakes, does the task in an isolated environment, records what it did, and stops at its cap.

• Repository maintenance and dependency updates.
• Code review and pull-request drafting when a GitHub event arrives.
• Documentation generated from the current source tree.
• Backlog items turned into draft changes.

When managed Codex is the right fit, and when to self-host

Managed hosting fits when you want a Codex agent working around the clock without becoming its on-call engineer. Provisioning, isolation, uptime, and monitoring are handled, and you spend your time on what the agent does, not on the box it runs on. Self-hosting still makes sense when you already operate servers and one more process is marginal, or when a task genuinely needs host-level control or an air-gapped network. The honest rule is the same as for any runtime: if the operational hours are already spent, self-host; if they would be new, they are the real cost.