OpenClaw ChatGPT Subscription: Codex Setup Guide (2026)

Nathan FieldNathan FieldFounder, Havstock8 min read
Illustration of a setup guide titled Connect OpenAI to OpenClaw listing What You Need and Verify It Is Working

Yes, you can run OpenClaw on your ChatGPT subscription instead of paying per API token. Log in with your ChatGPT account through OpenClaw's OpenAI provider, set openai/gpt-6-astra as your default model, and the Codex limits on your plan cover the work. ChatGPT Plus at $20 a month is enough to start.

Updated 29 September 2026. The first version of this guide (February 2026) used a provider called openai-codex and the model GPT-5.3 Codex. OpenClaw has since folded that into its main openai provider, and the old login name no longer works for new setups. Every command below was re-checked against OpenClaw's docs on 29 September 2026.

The short answer

  1. Log in: openclaw models auth login --provider openai
  2. Set the model: openclaw config set agents.defaults.model.primary openai/gpt-6-astra
  3. Check it: openclaw config get agents.defaults.model --json

No API key. No per-token bill. The usage comes out of your ChatGPT plan.

What "OpenClaw Codex" means

A lot of people search "OpenClaw Codex". That is because the ChatGPT route runs on OpenAI's Codex, and the naming is confusing.

Codex is OpenAI's agent engine. It is what runs the Codex app and the Codex command line tool. OpenClaw has an official codex plugin, and OpenClaw's docs describe it like this: it "runs embedded OpenAI agent turns through Codex app-server instead of the built-in OpenClaw harness."

In plain words: OpenClaw still runs your chat channels, sessions, tools and approvals. Codex does the thinking. Your ChatGPT plan pays for it.

Why ChatGPT is the easy subscription for OpenClaw

OpenClaw is built for it. The OpenAI setup page walks you through a ChatGPT login step by step. OpenClaw's OpenAI docs also say "OpenAI explicitly supports subscription OAuth usage in external tools and workflows like OpenClaw." We could not find that line on an OpenAI page, so treat it as OpenClaw's word, not OpenAI's.

The two projects are also close. OpenClaw's founder, Peter Steinberger, joined OpenAI in February 2026. The project is now run by the OpenClaw Foundation, an independent non-profit. Its README is clear on the money side: "OpenAI is a donor, not an owner."

Claude is a different story. We cover that further down.

The ChatGPT route is the one OpenClaw documents first. That makes it the safest subscription to build on today.

What You Need Before You Start

  • A ChatGPT plan that includes Codex. Plus ($20 a month) is the sensible start. The plan table is below.
  • OpenClaw installed and the gateway running. Update to the latest release first, because the old openai-codex setup is now a legacy route. Install steps are on the OpenClaw GitHub page.
  • A terminal: Mac Terminal, a Linux shell or Windows WSL.
  • A browser to finish the login. On a server with no screen, use the device code option in Step 1.

You do not need to reset your config. If OpenClaw already runs on another provider, those settings stay.

Step 1: Connect OpenAI to OpenClaw

New install? Run the setup wizard and pick OpenAI:

openclaw onboard --auth-choice openai

Already set up? Log in directly:

openclaw models auth login --provider openai

A browser window opens. Sign in to your ChatGPT account and approve the access. Then go back to your terminal.

Running OpenClaw on a VPS or a Mac mini with no screen? Add --device-code. You finish the sign-in on your phone or laptop instead of a local browser window.

openclaw models auth login --provider openai --device-code

Step 2: Set Codex as your default model

openclaw config set agents.defaults.model.primary openai/gpt-6-astra

OpenClaw's setup page says a fresh subscription setup uses openai/gpt-6-astra. If your account does not offer Astra, pick a model it does offer. This shows what your plan gives you:

openclaw models list --provider openai

One date to watch. OpenAI's Codex pricing page says GPT-5.5 "retires from ChatGPT, ChatGPT Work, and Codex on all plans on October 14, 2026." If your config still points at a GPT-5.5 model, move it before then.

Step 3: Verify It Is Working

openclaw config get agents.defaults.model --json

You should see openai/gpt-6-astra, or the model you picked, as the primary. Then send your agent a test message in any channel. If it answers, your ChatGPT plan is now paying for OpenClaw.

If the login failed, run Step 1 again. It is safe to repeat.

Moving from the old openai-codex setup

If you followed the February version of this guide, your config uses openai-codex/gpt-5.3-codex. OpenClaw now treats that as legacy. Its authentication docs say "don't create new openai-codex profiles" and point you to one command:

openclaw doctor --fix

That rewrites your old openai-codex entries to the new openai route. Then run Step 2 and Step 3. Do it from Terminal, not from inside OpenClaw. The rule below explains why.

Which ChatGPT plan should you use with OpenClaw?

Every ChatGPT plan, even Free, includes some Codex use. The difference is how much you get before you hit the limit.

PlanPrice (US)What OpenAI says it is forMessages per 5 hours on GPT-6 Astra
Free$0"Explore Codex capabilities on quick coding tasks"Not listed
Go$8 a month"Use Codex for lightweight coding tasks"Not listed
Plus$20 a month"Power a few focused coding sessions each week"5 to 45
ProFrom $100 a month"5x or 20x more Codex usage than Plus"25 to 225 (5x), 100 to 900 (20x)
Business$20 per user a month, billed yearly, 2+ users"Bring Codex into your startup or growing business"5 to 45

Source: OpenAI's Codex pricing page, checked 29 September 2026. OpenAI adds that "Weekly limits may also apply."

Our take: start on Plus and watch the limit for a week. If you keep hitting the 5-hour cap, try Pro. Astra is the heaviest model. The same page lists lighter GPT-6 models, Sol and Luna, with far higher limits. For simple jobs your agent repeats all day, a lighter model stretches Plus a long way.

Can you use a Claude subscription with OpenClaw?

Technically, yes. Whether you should is less clear, and Anthropic's line has moved more than once in 2026.

Here is where it stands on 29 September 2026:

  • Usage still counts against your plan. Anthropic's Agent SDK support page (updated 16 June 2026) says: "For now, nothing has changed: Claude Agent SDK, claude -p, and third-party app usage still draw from your subscription's usage limits." Anthropic had planned to move that use onto a separate credit, then paused the change.
  • The rules still limit tools like OpenClaw. Anthropic's Claude Code legal page says it "does not permit third-party developers to offer Claude.ai login into their own applications, or to route requests through Free, Pro, or Max plan credentials on behalf of their users." It also says it may enforce this "without prior notice."
  • OpenClaw's own advice is an API key. OpenClaw can reuse a Claude Code login on the same machine. But its Anthropic page says: "For shared production automation, use an Anthropic API key instead of Claude CLI."

Our take: if OpenClaw runs part of your business, do not build it on a Claude subscription. Use a ChatGPT plan, or pay for Claude through an API key so the bill is predictable and the setup is clearly allowed. If you are weighing Anthropic's own agent product instead, we compared the two in Claude Managed Agents vs OpenClaw.

A setup that can break without notice is fine for a side project. It is not fine for the thing that answers your leads.

The Self-Surgery Rule: Why You Should Never Configure OpenClaw From Inside OpenClaw

Here is a rule we follow without exception: change OpenClaw's setup from Terminal or VS Code. Never ask OpenClaw to change itself.

The reason is simple. If OpenClaw edits its own model, login or gateway settings while it is running, it is changing the system it runs on. If something breaks halfway, you can lose the agent and the tool you were using to fix it at the same time.

Terminal keeps you outside the system. You make the change, save it, and OpenClaw loads it cleanly on the next start.

That goes for every command in this guide, including openclaw doctor --fix. Run them all from Terminal.

FAQ

Can I use my ChatGPT subscription with OpenClaw?

Yes. OpenClaw supports ChatGPT sign-in through its openai provider. Run openclaw models auth login --provider openai, sign in, and set openai/gpt-6-astra as your default. The usage comes out of your ChatGPT plan's Codex limits instead of a per-token API bill.

What is OpenClaw Codex?

It is OpenClaw running on OpenAI's Codex engine. OpenClaw's official codex plugin hands each agent turn to Codex, while OpenClaw keeps your channels, sessions and approvals. Your ChatGPT plan pays for the Codex part. It is not a separate product you buy.

Why does openai-codex not work any more?

OpenClaw merged it into the main openai provider. Its docs now call openai-codex legacy and say not to create new profiles with it. If your config still has openai-codex/gpt-5.3-codex, run openclaw doctor --fix from Terminal, then set a current model.

Which model should I use with my ChatGPT subscription?

OpenClaw's setup page says a fresh subscription setup uses openai/gpt-6-astra. If your plan does not offer Astra, run openclaw models list --provider openai and pick one it does. Avoid GPT-5.5 for anything long term: OpenAI retires it from Codex on 14 October 2026.

Is ChatGPT Plus enough for OpenClaw?

For one agent doing everyday tasks, start there. Plus is $20 a month and OpenAI estimates 5 to 45 messages per 5 hours on GPT-6 Astra, with weekly limits possible. Lighter models get far more. If you keep hitting the cap, Pro starts at $100 a month with 5x the usage.

Can I use a Claude subscription with OpenClaw?

It works technically, and Anthropic's June 2026 support page says third-party app use still draws from your plan's limits. But Anthropic's rules limit routing plan credentials through other tools, and it may enforce that without notice. OpenClaw itself recommends an Anthropic API key for production use.

Do I lose anything moving OpenClaw from Claude to ChatGPT?

No. OpenClaw's README says you can swap model providers "without changing anything else." Your channels, skills and gateway stay the same. Only the model underneath changes. Run openclaw models list --provider openai after the switch to confirm the new model, and you can switch back later if Anthropic's rules settle.


If your OpenClaw agent works inside GoHighLevel (we may earn a commission from this link), our OpenClaw and GoHighLevel guide shows how to connect the two. And if you would rather have the funnels and follow-up around it built and run for you, see how Havstock Unlimited works. One small team handles it for one flat monthly fee.

GoHighLevel

Free trial
30 days free14 days
Included
Missed call text backCRM and pipelinesTexts and email in one inboxCalendars and bookingFunnels and websitesWorkflows and automationsPhone numbers and callingFree live bootcamp
Start free trialSigning up direct gets 14 days.We may earn a commission.
Talk to Nate, Taylor and Channa

You run the business.
We’ll grow it.

Tell us what you need. We will call you to plan it.

GoHighLevel free for 30 daysPlus a free live bootcamp
Start trial