Connect your ChatGPT subscription

If you pay for ChatGPT Plus, Pro, or a Team or Enterprise seat, that plan includes OpenAI model usage that you can spend through Codex. connect0 can use the same allowance for your own agent runs. Calls that run on your plan cost you nothing for the model tokens, and connect0 charges no fee on them up to an included amount each calendar month (the allowance is yours, wherever you use it); past that, only connect0's small per-token orchestration fee applies, charged to the account the agent ran in — never a charge for the tokens themselves. The current allowance and fee are listed below under What is billed.

This feature is rolled out gradually. If you do not see Settings → Providers yet, it is not switched on for your workspace.

Before you start

• A ChatGPT Plus, Pro, Team or Enterprise plan. The free tier does not include enough Codex usage to be useful here, and connect0 does not offer this for it.
• Device code authorization turned on in ChatGPT. As of September 2026 the setting is at ChatGPT → Settings → Security → "Enable device code authorization for Codex" (personal accounts). If your ChatGPT account belongs to a Team, Business, Enterprise or Edu workspace, this is a workspace permission that starts off, and a workspace owner or admin has to turn it on in the workspace's permission settings before you can continue.
• You must be signed in to connect0 as yourself. Agents cannot do this step for you.

The three steps

1. Turn on device code authorization. In connect0, open Settings → Providers → Connect ChatGPT (or the Connect ChatGPT card on your getting-started page) and click Open ChatGPT settings. In ChatGPT, go to Settings → Security and turn on Enable device code authorization for Codex. Come back to connect0.
2. Open the OpenAI device authorization page. Click Open device authorization. It opens https://auth.openai.com/codex/device in a new tab. Sign in there with the same OpenAI account that holds your ChatGPT plan.
3. Enter the code. connect0 shows a short code such as 5KNY-69NGB. Type it — or paste it with the copy button — into the OpenAI page and approve. Codes expire after 15 minutes.

What "waiting for approval" means

After step 2 the connect0 page says Waiting for approval — keep this page open. connect0 asks OpenAI every few seconds whether you have approved the code. As soon as you approve it in the OpenAI tab, the page changes to Connected as ‹your email› (ChatGPT ‹plan›) and you are done. You can close the OpenAI tab. If you close the connect0 page first, the code is not lost — reopen Settings → Providers within 15 minutes and it is still waiting.

Prefer the terminal?

From any MCP client connected to connect0 (Claude Code, Cursor, or your own), ask it to run the tool connect0_chatgpt_connect. It returns the same code and the two links. Then run connect0_chatgpt_status to see whether the code was approved.

Or with curl, using a connect0 access token for your user:

curl -X POST https://api.connect0.ai/v1/me/providers/chatgpt/device-code \
  -H "Authorization: Bearer $CONNECT0_TOKEN"
# → { "id": "…", "user_code": "5KNY-69NGB", "verification_url": "https://auth.openai.com/codex/device", "interval_seconds": 5, "expires_at": "…" }

curl https://api.connect0.ai/v1/me/providers/chatgpt/device-code/$ID \
  -H "Authorization: Bearer $CONNECT0_TOKEN"
# → { "status": "pending" }   …   { "status": "approved", "plan_type": "plus", "email": "…" }

What runs on your plan

• Which models. OpenAI models that your plan covers in Codex. In your model picker they carry a may run on your plan badge — "may", because which models a given plan can call is OpenAI's decision at request time; everything else runs on credits or your account's own OpenAI key as before.
• Which runs. Only runs that you start — a chat message, a "run now", a request from Zero or from an MCP client signed in as you. Scheduled runs, webhook-triggered runs, and runs a teammate starts on a shared agent never use your plan, even if the agent is yours.
• Your limits are your plan's limits. OpenAI meters Codex usage on a rolling five-hour window and a weekly allowance that depends on your plan, and it counts everything you do in Codex, ChatGPT and connect0 together. Settings → Providers shows how much of each window is used and when it resets. When a window is exhausted, the run stops and tells you when it resets. If you would rather keep going on account credits in that case, turn on Fall back to account credits when my plan is exhausted in Settings → Providers.
• What is billed. Calls that run on your plan use no connect0 credits for the model tokens. They still appear in Usage and in run explanations, marked ChatGPT plan, at zero cost while you are within your included allowance and at the orchestration fee after it, with the remaining allowance shown. Everything else in the run — connectors, tools, storage — is billed exactly as it always is. Your ChatGPT subscription is billed by OpenAI, not by connect0.

The included allowance and the fee

	Amount
Included per person, per calendar month (UTC)	2,000,000 tokens on your plan, in and out combined, with no connect0 fee
Past the allowance	US$0.50 per 1,000,000 tokens orchestration fee, charged in credits to the account the agent ran in
Model tokens themselves	never charged by connect0 — they are your plan's

The allowance counts every call that ran on your plan across every account you are a member of. Settings → Providers shows how much is left, updated hourly.

If something goes wrong

What you see	What it means	What to do
The OpenAI page says device code authorization is not enabled, or asks you to contact your workspace admin	Step 1 was skipped, or your workspace has not enabled it	Personal account: turn on the setting under Settings → Security and get a new code. Workspace account: ask a workspace owner or admin to enable device code authentication, then try again.
That code expired	More than 15 minutes passed	Click Get a new code.
OpenAI didn't approve that code	The approval was declined, or the code was entered on an account where the setting is off	Check the setting, get a new code, and approve it while signed in to the right OpenAI account.
Connected, but the plan shows as Free	You approved with an OpenAI account that has no paid plan	Disconnect, then connect again signed in to the account that holds your subscription.
Your ChatGPT connection stopped working in Settings	OpenAI ended the grant — you turned the setting off, changed your password, revoked it in ChatGPT, or OpenAI withdrew it	Click Reconnect. If it keeps happening, check the setting in ChatGPT.
A run stops with Your ChatGPT plan's 5-hour window is used up. Resets at ‹time›	You hit your plan's limit	Wait, or turn on the fallback preference, or click Continue on account credits for this run for that one run.
Your agent used credits for an OpenAI model you expected to be free	The run was not started by you, or the model is not covered by your plan, or your preference Use my ChatGPT plan for eligible models is off	Check the run's explanation — it names which credential each call used.

Disconnect

Settings → Providers → ChatGPT → Disconnect. connect0 deletes the tokens it holds at once. Your agents go back to account credits or your account's own OpenAI key for OpenAI models. Nothing changes on your ChatGPT account, and you can turn the device code setting off in ChatGPT whenever you like.

Good to know

connect0 is not affiliated with OpenAI. Using your ChatGPT plan in connect0 is subject to OpenAI's terms of use, and OpenAI may change or withdraw the ability to do this at any time. connect0 identifies itself to OpenAI as connect0 on every request and never pretends to be another product.

Where to go next

• Bring your own key — the account-level alternative: your provider contract, for every run on the account.
• Choosing a model — where the may run on your plan badge shows up.
• Credits and pricing — how the orchestration fee is charged.