You're already paying twenty bucks a month for ChatGPT Plus.
You're also paying OpenAI per-token for your agent's API calls.
Notice anything weird about that?
The thing nobody mentions
The Codex models that live inside ChatGPT — GPT-5.5, GPT-5.4, GPT-5.4-mini — are the same models OpenAI sells through the API. They're just billed differently: monthly subscription instead of per-token. And OpenAI ships an official OAuth flow that lets external tools (yes, including your ClawdBot agent) use your subscription's quota directly.
We just wired it into your dashboard. Your agent can now sign into ChatGPT the same way you do — and stop running up an API bill while it works.
What you get
A new row in your Keys panel: ChatGPT (Plus, Pro, or Team) with a little sparkle next to it. Click connect, sign in once, and your agent has access to the same Codex models you get inside chatgpt.com:
- GPT-5.5 — flagship
- GPT-5.4 — workhorse
- GPT-5.4-mini — fast and cheap (in compute terms; for you it's all free)
- GPT-5.3-Codex — purpose-built for coding tasks
Same model. Same quality. Same speed. Different bill.
What it costs
Nothing extra. You're paying the subscription anyway.
A single API request to GPT-5.5 through OpenAI's API runs you ~$0.125 per 1K input tokens and ~$0.75 per 1K output. A complex agent task can burn through dollars in minutes. Through your $20 Plus subscription, the same task costs $0. Through Pro at $100, you get roughly 5× the throughput. Same models, the meter just doesn't run.
The limits per plan
Every plan has a rolling 5-hour message cap. Here's the official rate card:
| Plan | Monthly | GPT-5.5 messages / 5h | GPT-5.4 messages / 5h |
|---|---|---|---|
| Plus | $20 | 15 – 80 | 20 – 100 |
| Pro 5× | $100 | 80 – 400 | 100 – 500 |
| Pro 20× | $200 | 300 – 1,600 | 400 – 2,000 |
| Business | per-seat or PAYG | token-based credits | token-based credits |
The ranges depend on task complexity — short prompts cost less of your quota than multi-turn agentic chains.
Will Plus be enough for me?
Before you connect, do this 30-second math.
What "one message" actually costs
The 80-per-5-hours cap is abstract until you know what a "message" looks like in real agent work:
| Your agent does this | ~Messages used |
|---|---|
| Replies to a quick chat | 1 – 3 |
| Replies and uses a tool (lookup, email, fetch a page) | 3 – 8 |
| Runs an agentic chain — "schedule the meeting, send confirmation, log to CRM" | 10 – 30 |
| Fires a background loop (memory consolidation, scheduled cron) | 5 – 15 each |
| Handles a voice-call turn (transcribe → reason → respond → speak) | 5 – 12 |
One "book me a flight to NYC next Tuesday" interaction can quietly eat 20 of your 80 messages. The cap feels different now.
Match the plan to how you'll use it
| If your agent is mostly… | Pick |
|---|---|
| You tinkering, building skills, learning | Plus ($20) alone |
| A personal assistant — calendar, email, daily summaries | Plus ($20) + Anthropic or OpenAI key as a fallback |
| A real workflow for one operator's business | Pro ($100) |
| Multi-user, customer-facing, or with the voice gateway active | Pro $200 or Business per-seat |
| High-traffic production where rate-limit blocks are unacceptable | API keys only — pay per token, no caps |
Two takeaways:
- Light use? Plus is genuinely free money. Connect it.
- Anything serious? Plus alone will hit the cap at the worst possible moment. Either pick a tier that matches your throughput, or keep an Anthropic/OpenAI key in the Keys panel as a safety net — your agent uses Codex first, then falls back automatically when capped.
Two more things worth knowing
Codex shares its quota pool with other agentic features. Right now that's ChatGPT for Excel; OpenAI is adding "Workspace Agents" next. Heavy Excel use eats your agent's budget — same bucket.
Check your live usage anytime at chatgpt.com/codex/settings/usage — your current 5-hour burn, weekly cap, remaining credits, reset time. Bookmark it.
How to connect (2-minute walkthrough)
Open your dashboard, expand Keys on the agent you want to connect, scroll to the ChatGPT row.
1. Click Sign in with ChatGPT.
2. A modal opens. Click the big orange Open ChatGPT sign-in button. New tab.
3. Sign in with the ChatGPT account whose subscription you want to use.
4. This is the part that looks broken but isn't. Your browser will redirect to a page that says "Hmmm... can't reach this page. localhost refused to connect." That's expected. The redirect target is on your agent's server, not your laptop. Don't refresh. Don't panic.
5. Copy the entire URL from the address bar. (It starts with http://localhost:1455/auth/callback?code=...)
6. Switch back to your dashboard tab. The modal is still open with a paste field. Paste the URL in.
7. Click Finish sign-in.
8. The modal flips through "Exchanging tokens..." for a few seconds, then shows ✓ Success and auto-closes.
9. The row in your Keys panel is now green, showing your email and plan tier (e.g. you@example.com · Plus).
That's it. Your agent now uses your subscription for Codex requests automatically. No code change, no key to copy around, no .env file to edit.
How the bridge actually works (skip if you don't care)
OpenAI's OAuth flow always redirects to localhost:1455. That URL is baked into the upstream Codex CLI for security reasons — they don't want random public domains in OAuth redirect URIs. But your agent runs on a server, not your laptop, so the redirect can never reach you directly. The workaround:
- We open OpenAI's auth URL in your browser.
- After you sign in, OpenAI redirects with a single-use code in the URL (which fails to load — that's the "Can't reach this page").
- You paste that URL back into our modal.
- We send the URL to your agent through your secure dashboard channel.
- Your agent's CLI is listening on its OWN
localhost:1455(on the agent's machine), receives the URL, exchanges the code for tokens, and stores them locally.
We never see your access token, refresh token, or any conversation content. The only thing that briefly transits our infrastructure is the one-time OAuth code (single-use, 10-minute TTL). Your tokens live exclusively on your agent's host, refreshed automatically by the Codex CLI when they expire.
How to disconnect
Same row. Click Disconnect. Confirm in the modal. Your agent stops using the subscription within about 10 seconds, the tokens get wiped from disk, and the row goes gray. Your ChatGPT account is unaffected — you can still use chatgpt.com normally; you've just revoked one third-party app's access.
Useful links
- OpenAI: Using Codex with your ChatGPT plan — the official policy
- OpenAI: Codex pricing — the rate card with all the numbers
- OpenAI: Codex authentication docs — how OAuth works under the hood
- Your Codex usage dashboard — bookmark this
- OpenAI: Codex CLI on GitHub — the upstream tool whose OAuth flow we use
Get started
If you're paying for ChatGPT, you've already done the hard part. Open your dashboard, find the ChatGPT (Plus, Pro, or Team) row, click connect.
Two minutes from now, your agent will be running on the subscription you already pay for — and your API bill stops growing while it works.