Skip to content

Provider ChatGPT Codex

wy1024 edited this page Mar 31, 2026 · 1 revision

ChatGPT (Codex) ★ Recommended

Use your existing ChatGPT Plus or Pro subscription to power your OpenClaw instances — flat-rate, no per-token billing.

Overview

Detail Value
Provider name in ClawFleet openai-codex
Authentication OAuth (ChatGPT login)
Preset models gpt-5.4, o3, gpt-5-mini
Pricing Included in ChatGPT Plus ($20/mo) or Pro ($200/mo)

Why recommended?

  • No API key needed — log in with your ChatGPT account
  • Flat-rate billing — your existing subscription covers inference, no per-token charges
  • Same models as OpenAI API — GPT-5.4, o3, and more
  • Lowest onboarding friction — most users already have a ChatGPT subscription

Step 1: Have a ChatGPT subscription

You need an active ChatGPT Plus ($20/month) or ChatGPT Pro ($200/month) subscription. Free-tier ChatGPT accounts do not have API access.

Step 2: Add in Dashboard

  1. Open the ClawFleet Dashboard (http://localhost:8080)
  2. In the sidebar, click Assets > Model Config
  3. Click + Add Model Config
  4. Fill in the form:
    • Name: e.g. "ChatGPT GPT-5.4"
    • Provider: select ChatGPT (Codex) ★ Recommended (selected by default)
    • Model: select gpt-5.4 (recommended), o3, or gpt-5-mini
  5. Click Login with ChatGPT
  6. A browser window opens — log in with your ChatGPT account and authorize
  7. The window closes automatically and Dashboard shows "ChatGPT connected"
  8. The model config is saved automatically

No API key field is shown — authentication is handled entirely through the OAuth login.

Step 3: Assign to an instance

  1. Go to Fleet > Instances
  2. Click Configure on the instance card
  3. Select your ChatGPT model config from the Model Config dropdown
  4. Optionally select a channel config
  5. Click Configure

Pricing notes

  • ChatGPT Plus ($20/month) includes a rolling usage window — sufficient for moderate usage
  • ChatGPT Pro ($200/month) provides near-unlimited inference
  • No separate API billing, no credit purchases, no pay-per-token anxiety
  • Compare: OpenAI API charges $2.50/MTok input for GPT-5.4

Remote Dashboard (SSH tunnel)

When accessing the Dashboard remotely via SSH tunnel, include port 1455 for the OAuth callback:

ssh -L 8081:localhost:8080 -L 1455:localhost:1455 -fN user@server

Then open http://localhost:8081 in your browser. The OAuth flow works transparently.

Troubleshooting

"Login with ChatGPT" opens a blank page or fails

  • Ensure port 1455 is not blocked by a firewall
  • If using a remote Dashboard, verify the SSH tunnel includes -L 1455:localhost:1455
  • Try closing and reopening the Dashboard

OAuth succeeds but instance won't respond

  • Ensure the instance is running (check status on the instance card)
  • Check that your ChatGPT subscription is active (expired subscription = no inference)
  • Try a different model (e.g. gpt-5-mini for faster response)

"Token exchange failed"

  • This usually means the OAuth session expired — click "Login with ChatGPT" again
  • Ensure you completed the login in the popup window before it timed out (5 minutes)

Clone this wiki locally