Can you use a Claude or ChatGPT subscription for automation? What Anthropic and OpenAI say (October 2026)

Yes, but the two companies say different things. As of 25 October 2026, Anthropic says Claude Code scheduled tasks, claude -p and third-party apps all draw from your Claude plan's limits. OpenAI says an API key is "the right way to authenticate automation", and lets only listed partner apps use your ChatGPT plan.

Rules like these change. Anthropic announced a change to subscription use by the Claude Agent SDK, then paused it on 15 June 2026. Every statement below carries the date its source says it applies, and a link to that source, so you can check whether it still holds.

What does Anthropic say about automating with a Claude plan?

Anthropic's help article Use the Claude Agent SDK with your Claude plan (article 15036540) carries two dated updates:

• 15 June 2026: "We've paused the previously-announced changes to Claude Agent SDK usage. For now, nothing has changed: Claude Agent SDK, claude -p, and third-party app usage still draw from your subscription limits."
• 7 October 2026: "Claude Max and Team plans now include monthly API credits, which cover the Claude Agent SDK, claude -p, the Claude API, and Claude Managed Agents. You can still use the Claude Agent SDK, claude -p, and third-party apps with your subscription limits."

Anthropic also ships its own automation on the plan. The Claude Code docs say cloud scheduled tasks are available on Pro, Max, Team and Enterprise, run on Anthropic's cloud "so they keep working when your laptop is closed", and "draw down subscription usage the same way interactive sessions do". The feature is labelled research preview.

So, as of October 2026, Anthropic's own pages say a Claude plan can pay for unattended work: its own scheduled tasks, your own claude -p scripts, and third-party apps.

What does OpenAI say about automating with a ChatGPT plan?

OpenAI draws the line differently. Its Codex docs on non-interactive mode and keeping Codex account auth in CI/CD say:

• "The right way to authenticate automation is with an API key. Use this guide only if you specifically need to run the workflow as your Codex account."
• Running codex on a runner signed in with your ChatGPT account is documented as "an advanced workflow for enterprise and other trusted private automation". The same page says not to use it for public or open-source repositories.
• "When you sign in with an API key, Codex uses standard API pricing instead of included ChatGPT plan credits" (Authentication).

OpenAI's own scheduled tasks do run on the plan. Scheduled tasks in ChatGPT and Codex can run on a schedule or from Gmail, Slack and GitHub events on eligible plans. Project-scoped tasks need "the machine powered on and the ChatGPT desktop app running".

For other companies' apps, OpenAI runs Sign in with ChatGPT. In listed partner apps, "eligible ChatGPT Plus and Pro subscribers can also choose to use their ChatGPT plan for AI requests", and that use "counts toward your existing plan limits". On 25 October 2026 the partners with plan usage were:

• Amp Code, Conductor, Dactyl, Devin, Hermes Agent, Hyperagent
• Kilo Code, Notion, Vercel, Vorflux, Warp
• Lovable (coming soon)

How do the two compare, side by side?

Question	Anthropic (Claude plan)	OpenAI (ChatGPT plan)
First-party scheduled runs on the plan	Yes: cloud scheduled tasks, research preview	Yes: scheduled tasks in ChatGPT and Codex
Your own headless scripts on the plan	claude -p draws from plan limits (15 Jun and 7 Oct 2026 notes)	Documented as advanced; API key recommended
Third-party apps on the plan	"Still draw from your subscription limits" (7 Oct 2026)	Only Sign in with ChatGPT partners, listed by OpenAI
Recommended credential for CI	Not stated in the sources above	API key
Separate credits for automation	Max and Team monthly API credits (7 Oct 2026)	Credits can extend eligible usage (plan dependent)
Source	Article 15036540, scheduled tasks docs	Auth, CI/CD auth, Sign in with ChatGPT

What does this mean for a tool that runs your plan for you?

Hosted tools that run Claude Code or Codex for you fall under these same rules, and the two sides differ.

On the Claude side, Anthropic's 7 October note covers third-party apps drawing from your subscription limits. That includes services such as Kavela, which runs the Claude Code CLI in a cloud computer signed in to your Claude plan (Kavela docs).

On the OpenAI side, the picture is less settled. Kavela signs the Codex CLI in to your ChatGPT account with OpenAI's device sign-in page, and Kavela's docs say your plan then pays for the model. Kavela is not on OpenAI's Sign in with ChatGPT partner list. OpenAI has published no statement we could find about a third-party service running the Codex CLI on a user's ChatGPT sign-in. Treat the ChatGPT side as unconfirmed by OpenAI until it is, and decide for yourself whether an API key is the safer choice for your workflow.

How do you check where your usage goes?

1. Claude: open claude.ai/settings/usage. Anthropic's scheduled-task docs point there for current consumption. Max limits reset every five hours, with a weekly limit across all models (What is the Max plan?).
2. ChatGPT: open Settings, then Usage. OpenAI's Sign in with ChatGPT page says connected apps show there, and you can set each app's weekly limit as a percentage of your plan.
3. Before you schedule anything, run the prompt once by hand and read the result. OpenAI's scheduled-task docs say to "test its prompt in a regular web chat" first, and the same habit saves Claude usage too.
4. Write the date you checked these rules into your runbook, next to the links above.

What happens when automation hits the plan limit?

It depends on where the job runs:

• Claude Code scheduled tasks have their own hourly caps on top of plan usage, such as 100 scheduled runs per hour per account, and "none of these hourly limits has overage" (scheduled tasks docs).
• Your own claude -p or codex exec script fails or waits, depending on how you wrote it. A cron job with no retry simply loses that run.
• Kavela: the task waits and continues after your plan resets. It does not switch to paid credits on its own (Kavela docs).

How do you set this up on Kavela?

Connect your Claude or Codex plan once for the workspace, then paste a prompt like this into Kav's chat box and replace the parts in angle brackets:

Every weekday at 09:00 <your timezone>, check <what to check: a GitHub repo, a Notion page, a website> and write me a short report of what changed since the last run.
Run it on my connected Claude plan. If the plan reaches its limit, wait for the reset rather than using credits.
Hold each report to this checklist: <2 or 3 things a good report always has>.
Send the report to my approvals first; after I approve the first three, post it to <Slack channel or email address>.
Connect <GitHub / Notion / Slack> when you need it.
Run it once now so I can check the first result, then set the routine.

You will still sign in to each service the prompt names, and you approve the first results. Setup takes about 10 minutes if you already have those accounts. For how scheduled work runs on Kavela, read schedule a routine. For the other ways to keep Claude Code running, see how to run Claude Code 24/7.

Frequently asked questions

Is it against Anthropic's terms to automate Claude Code on a Pro or Max plan?

Anthropic's own help article, updated 7 October 2026, says the Agent SDK, claude -p and third-party apps "still draw from your subscription limits", and Anthropic sells cloud scheduled tasks on Pro and Max. Read article 15036540 again before you rely on it; Anthropic has changed this once already.

Can a third-party app use my ChatGPT Plus or Pro plan?

Only apps in OpenAI's Sign in with ChatGPT program, where you approve "Use your ChatGPT plan" on a consent screen. OpenAI's Codex docs recommend an API key for other automation.

Do the Claude Max monthly API credits cover Claude Code?

No. They cover the Agent SDK, claude -p, the Claude API and Managed Agents (article 15036540, 7 October 2026). Anthropic's Claude Code help article says they cannot be used in Claude Code or the Claude apps.

Why does the Codex side say unconfirmed?

OpenAI documents two sanctioned routes for automation: an API key, and partner apps in Sign in with ChatGPT. A service signing the Codex CLI in to your account is neither. We found no OpenAI statement for or against it, so we say so instead of guessing.

Where can I compare the hosted options?

See run Claude Code or Codex in the cloud on your own subscription, which compares the first-party and third-party options side by side.

Stuck on setup? Ask in the Kavela Discord.