Using your ChatGPT subscription with OpenClaw (Codex sign-in)
Connect ChatGPT to OpenClaw, choose a model your account supports, and check what happens at usage limits. Subscription access and API billing are separate.
Yes. OpenClaw supports signing in with a ChatGPT account to use its Codex model allowance instead of a separately billed API key. Use OpenClaw’s own OpenAI sign-in, then select a model your account can use. Free and paid accounts have different allowances; a successful login is not a promise of unlimited usage or access to every model.
What your ChatGPT account pays for
OpenAI’s Codex pricing lists Free, Go, Plus, Pro and workspace options. The available usage depends on the plan and the work being done. OpenClaw’s OpenAI setup documentation provides a subscription sign-in alongside the separate API-key route.
This changes how you pay for supported model calls, not where OpenClaw runs. Your computer or hosting subscription is still needed. For the full budget, separate hosting, model and connected-tool costs. Paying for ChatGPT also does not add credit to an OpenAI Platform API account.
Connect the account to OpenClaw
Run this on the machine where OpenClaw runs, then complete the browser sign-in:
openclaw models auth login --provider openaiFor a server without a local browser, or a setup where the browser cannot return to the local callback, use the device-code option:
openclaw models auth login --provider openai --device-codeEnable device-code authorization first, then open the printed URL in your own browser and enter the code. A workspace account may need an admin to enable that permission. Use the account whose allowance you intend to spend.
Signing into Codex CLI separately is not the same as completing OpenClaw’s sign-in. Current OpenClaw onboarding does not automatically import the CLI’s saved credentials. Use the OpenClaw command above for this setup.
Verify a working model, not just a successful login
The model catalog exposed to your account is what matters. Check the available OpenAI models:
openclaw models list --provider openaiIn current native Codex sessions, /codex models lists the catalog and /codex status checks the runtime. Select an available model rather than assuming the default is included in your plan. If the response says a model is unsupported, check the model selection and account entitlement before repeating the login.
Send a small test message in the conversation you actually use. A conversation with its own model selection can differ from the agent’s default. The model-switching guide covers changing the right scope; a saved setting alone does not establish that the next reply will use it. If the agent has stopped replying altogether, follow the not-responding triage guide.
Decide what happens when the allowance runs out
You can wait for the allowance to reset, or configure another authorized source of model access. OpenClaw’s Codex authentication documentation describes moving to the next configured OpenAI profile when a subscription is rate-limited. If that next profile uses an API key, subsequent usage can incur API charges. Configure a paid fallback only if that is the behavior you want.
Check usage in the account as well as the agent. OpenClaw keeps subscription allowances and Platform API billing separate. A subscription usage limit is not a broken connection, and signing in again does not replenish it.
Check the other tools separately
Model access does not settle the authentication for every feature. Memory embeddings, voice and other tools have their own supported providers and routes. Before removing an API key, check what else uses it. If memory search stops after switching to subscription access, check its embedding provider rather than changing a working chat-model login.
On an Everpod pod, memory search keeps working after you connect a ChatGPT subscription. You still choose the model and whether to use a fallback when that subscription reaches its allowance.