How to run Codex on a schedule in 2026: 9 ways compared

Disclosure: Enjoy is our product, and it runs Codex. We checked every way below against its maker’s own pages on November 5, 2026, and we say where another way is the better pick, including the ones built into the ChatGPT app.

The ways to run Codex on a schedule, compared

A scheduled run is Codex working without you: a prompt you save once, started at the times you choose, such as a report on yesterday’s commits every weekday at 9:00. The question that decides your pick is where it runs. Anything that needs the files on your computer needs that computer on and awake, and OpenAI says so: “Keep the computer on and the app running when a scheduled task needs local files.”

Way	Best for	Price, checked November 5, 2026	Where it runs	Your computer on?	Agents
Standalone scheduled tasks, in the ChatGPT app	Recurring work on your own projects	Listed on ChatGPT Plus ($20 a month), Pro (from $100), Business and Enterprise	Your computer	Yes, with the app running	Codex
Scheduled tasks inside a chat	Checking on something, in the same chat	The same plans	Your computer, for local chats	Yes, for local chats	Codex
Scheduled tasks on ChatGPT’s web	Reports from connected tools, not your folders	When enabled for your workspace	OpenAI’s cloud	No	ChatGPT Work
Dots	A cloud repository, with your laptop shut	Pro, Business Premium and Enterprise, rolling out gradually	OpenAI’s cloud, or a connected computer	Not for cloud tasks	Codex and ChatGPT Work
GitHub Actions	Scheduled jobs in a repository’s CI	An OpenAI API key, plus GitHub’s runner minutes, free in public repositories	GitHub’s runners	No	Codex
cron, with codex exec	A server that’s always on	Free; runs on your ChatGPT plan or an API key	Any computer you run	Yes	Codex, or any command-line agent
Paseo	Many agents on schedules, free	Free	The machine running its daemon	Yes, that machine	Codex, Claude Code, Copilot, OpenCode, Antigravity, Muse Code and more
Orca	Many agents on schedules, free	Free	Your computer, or an Orca server you run	Yes, that machine	Codex, Claude Code and any other coding agent
Enjoy	Any lab’s agent, across all your projects	A monthly subscription, free for 7 days, with your phone and other computers (pricing)	Your computers	Yes, with Enjoy open	Codex, Claude Code, Grok Build, Antigravity, Muse Code, OpenCode

Prices are per month as each maker lists them in the US, before tax. OpenAI’s pricing lists “Project and standalone scheduled tasks” on Plus, Pro, Business and Enterprise. Its Free and Go plans aren’t in that comparison, and it describes their Codex as “in the desktop app, subject to rollout.” Business is $20 a user a month billed yearly, for two or more users, or $25 billed monthly.

Standalone scheduled tasks, in the ChatGPT app

Best for: recurring work on the projects on your own computer, when Codex is your only agent.

Codex’s scheduler is in the ChatGPT desktop app, and it used to be called automations. Open Scheduled in the sidebar, or describe the job in a chat and say when it should run. Each run starts a new chat and reports back to Scheduled, which “acts as your inbox.” One task can cover more than one project, and in a Git repository it can run in your project or on its own worktree, which keeps its changes “separate from unfinished local work.” For an unusual schedule, you can edit its RFC 5545 recurrence rule, such as RRULE:FREQ=MONTHLY;BYMONTHDAY=1;BYHOUR=9;BYMINUTE=0.

Price: included in the plans above. The app runs on Mac and Windows, and “The ChatGPT desktop app for Linux is available in preview.”

Watch for: “For project-scoped scheduled tasks, keep the machine powered on and the ChatGPT desktop app running.” On OpenAI’s forum, one user’s task didn’t run while their laptop slept and started only once they woke it. OpenAI Support replied that for local files “the computer still needs to stay awake with the desktop app running.” The docs don’t say what happens to a time you miss. Tasks “run unattended and use your default sandbox settings,” and on full access “ChatGPT may change files, run commands, and access network without asking,” so OpenAI suggests workspace write instead. And “frequent schedules can create many worktrees over time,” so archive the runs you don’t need.

Scheduled tasks inside a chat

Best for: checking on a deploy, a pull request or a long job, with everything the chat already knows.

A task in a chat “uses the chat’s existing context instead of starting from a new prompt each time.” It can repeat every few minutes “for active follow-up loops,” or daily and weekly at a set time. OpenAI’s example is a skill that checks a pull request’s status with the GitHub plugin and fixes new review feedback. Ask in the chat, and say when to run and when to stop.

Price: the same as standalone tasks.

Watch for: OpenAI says to “make the prompt durable,” describing what to do on each run, how to tell whether anything’s worth reporting, and when to stop or ask you. A local chat has the same rule as any local task: the computer stays on with the app running.

Scheduled tasks on ChatGPT’s web

Best for: recurring reports from connected tools and uploaded files, with your computer off.

On the web, “create them from Chat or ChatGPT Work on the web and manage their runs from Scheduled.” They can use “uploaded files, connected tools, skills, and plugins.” On eligible plans, a task can start from a Gmail, Slack or GitHub event instead of a time, such as activity on a pull request. Workspaces can also set up Team Tasks, which “run in the cloud with the team’s service account.”

Price: included where available: “Availability depends on your plan and workspace settings.”

Watch for: this isn’t Codex in your project. Web tasks “can’t work directly in a folder on your computer,” and “They don’t keep a local folder or worktree available between runs.” One task “can’t combine event triggers with a time-based schedule.”

Dots

Best for: a repository in a Codex cloud environment, on a schedule, with your laptop shut, if you have a dot.

A dot is OpenAI’s “always-on agent,” and it “lives in the cloud and has its own computer and browser.” Its docs say “Work that repeats at fixed times needs a saved schedule,” which you ask for in words, with a time zone and an end date. It can also “Create a task in a Codex cloud environment you have already set up. Your computer can be offline.” You review recurring work in Scheduled.

Price: Pro at $100, $200 or $500 a month, outside the European Economic Area, the UK and Switzerland, plus Business Premium, and Enterprise, where an admin has to turn dots on. “Tasks your dot starts or manages in Work or Codex count toward those products’ usage limits as usual.”

Watch for: “Dots are rolling out gradually. You may not see dots immediately, even if your plan is eligible.” OpenAI documents recurring tasks and cloud coding tasks separately, with no example joining the two, so ask your dot to “Confirm the schedule” and read the first run. We didn’t find a time-based schedule for a Codex cloud environment anywhere else in OpenAI’s docs. Local work through a dot still needs the computer “online with the ChatGPT app open.”

GitHub Actions

Best for: scheduled jobs on a repository, run by your CI, with no computer of yours involved.

OpenAI’s Codex GitHub Action, openai/codex-action@v1, “installs the Codex CLI, starts the Responses API proxy when you provide an API key, and runs codex exec under the permissions you specify.” OpenAI’s example runs on pull requests. For a schedule, give the same job GitHub’s schedule event and a cron expression.

Price: API prices for Codex, as the action takes an OpenAI key from a GitHub secret. GitHub’s billing page says standard runners are free in public repositories, and GitHub Free includes $12 of compute a month for private ones, at $0.006 a minute on a standard Linux runner.

Watch for: GitHub says the schedule “can be delayed during periods of high loads,” that “some queued jobs may be dropped,” and that “Scheduled workflows will only run on the default branch.” In a public repository, they’re “automatically disabled when no repository activity has occurred in 60 days.” OpenAI warns that “Codex has broad access on GitHub-hosted runners unless you restrict it,” and that you should sanitize prompt inputs “to avoid prompt injection.” On Windows runners you have to set safety-strategy: unsafe.

cron, with codex exec

Best for: a server or a computer that’s always on, with nothing between your scheduler and Codex.

The do-it-yourself way: have cron, launchd or Windows Task Scheduler start Codex without its interface. “Non-interactive mode lets you run Codex from scripts,” say OpenAI’s docs, and they name “scheduled jobs” among its uses. A run that can edit files looks like codex exec --sandbox workspace-write "<task>", and codex exec resume --last picks up the last run. To start work in OpenAI’s cloud instead, codex cloud exec “submits a task directly.”

Price: free. “codex exec reuses saved CLI authentication by default,” so it runs on the plan you signed in with, from Plus, or you can set CODEX_API_KEY for a run and pay API prices.

Watch for: everything around the run is yours to build, from logs to telling you when it’s done. “By default, codex exec runs in a read-only sandbox,” so a job that should change files needs a flag, and old scripts break: “--full-auto is no longer supported.” On CI runners, OpenAI says “API keys are the right default for automation.” ChatGPT sign-in there is an advanced path where you keep auth.json between runs and “Treat ~/.codex/auth.json like a password,” and it says “Do not use this workflow for public or open-source repositories.”

Paseo

Best for: scheduling Codex, Claude Code and other agents for free, on a machine you keep running.

Paseo is free and open source, under Apache 2.0 on GitHub. Its schedules start “a new agent for you on a cron cadence: at this time, run this prompt, in this repo, with these agent settings.” Its heartbeats send “a recurring prompt back into one existing agent,” much like a scheduled task in a chat. You set them up in the app, by asking in a chat, or with paseo schedule create on a headless box.

Price: free. “You need agent providers installed with your own credentials.”

Watch for: the machine running its daemon has to stay on, though Paseo’s homepage says you can run it “on a home lab, or a cloud machine.” Each run’s workspace and agent are archived when it ends, unless you turn off Archive on finish in the app: “The CLI and MCP schedule tools do not expose this option.”

Orca

Best for: scheduling Codex and other agents for free, with checks that skip a run when there’s nothing to do.

Orca is “Free and open source for macOS, Windows, and Linux,” under the MIT license. Its automations “run a prompt on a schedule,” created with orca automations create and --provider codex, or on its desktop Automations page. A schedule can be a preset such as weekdays, “plus cron expressions or RRULE strings,” and a precheck can “Skip scheduled work when a cheap shell probe fails,” such as when there’s no pull request to review.

Price: free. You plug in the plans you already have.

Watch for: an automation runs on the machine that hosts it, shown in the column for “where each automation is stored and scheduled.” To keep going while your laptop sleeps, you need a Remote Orca Server of your own: “Orca does not sell managed VPS hosting.”

Enjoy

Best for: scheduling Codex beside any other lab’s agent, across all your projects, without writing cron.

Enjoy is a desktop app for Mac, Windows and Linux that runs Codex, Claude Code, Grok Build, Antigravity, Muse Code, and local models through OpenCode, on the plans you already have. A workflow “is a set of instructions you save once and run whenever you need it, or on a schedule.” What it adds for scheduled work:

• Any agent, on a schedule. Set Repeat to every day, every weekday, every week, or a cron expression, in the time zone you choose, and pick the agent, model, effort and permissions its runs use. One workflow can run Codex and the next Claude Code or a local model.
• Each run, a conversation. Every run starts a new conversation that you can read, reply to and steer, at your desk or from your phone.
• Limits handled. Sign in to several ChatGPT accounts, and when one runs out, Enjoy moves to another or waits for the reset, then carries on.
• A computer that stays on. Scheduled runs happen on the computer that has the project, so a Mac mini at home can run them all night while you follow along from your laptop or phone.

Price: Enjoy is one monthly subscription, and the first seven days are free, with a card to start. It includes your phone and any browser and your other computers, and each teammate you invite needs a subscription of their own. See pricing.

Watch for: scheduled workflows run only while your computer is awake with Enjoy open: “Enjoy must be running,” and “The computer must also be awake.” “Missed times are skipped, not made up,” and when the previous run “is still working or waiting for your answer, that scheduled time is skipped.” It runs the Codex CLI, which OpenAI includes from Plus, so ChatGPT’s Free and Go plans won’t do. Enjoy isn’t open source, and it’s a monthly subscription on top of your ChatGPT plan once the free week ends.

When the ChatGPT app is enough

If Codex is the only agent you use and your computer is usually on, stay with what’s built in. Standalone tasks cover your projects, tasks inside a chat cover the afternoon you spend waiting on a pull request, and neither costs anything beyond your plan, from Plus. It’s the better pick over Enjoy for that reader. And with your computer off, its web tasks and dots can keep working, where Enjoy can’t.

Look further when you want Claude Code or another lab’s agent on the same schedules, a job in your CI, or a server you control. Claude Code has its own schedulers, which our guide compares. And plan for the limits: scheduled runs draw on your plan’s usage, and from December, ChatGPT and Codex share one allowance on Plus, Pro and Business.

How we compared

We read OpenAI’s docs on scheduled tasks, non-interactive mode, its GitHub Action, dots and pricing, GitHub’s docs on scheduled workflows and billing, Paseo’s and Orca’s sites, docs and license, and one thread on OpenAI’s forum on November 5, 2026, and every one is linked below. We quote each maker’s own words for its limits and leave out what it hasn’t documented. We didn’t run these setups ourselves for this guide. Scheduling in these apps changes often, so we recheck this guide against its sources at least once a month, and the date at the foot of the page says when we last did.