Use Your ChatGPT Plus Subscription with Hermes Agent

Running OpenClaw instead of Hermes? See the OpenClaw ChatGPT subscription guide.

How It Works

Hermes Agent supports the openai-codex provider, which routes requests through your ChatGPT subscription instead of the pay-per-token OpenAI API. You authenticate via a device-code OAuth flow, and Hermes uses your subscription's included quota — no API billing.

What You Need

• An active ChatGPT Plus ($20/mo) or ChatGPT Pro ($200/mo) subscription
• A deployed Hermes Agent instance on OpenClaw Launch

Step 1: Open the Terminal

1. Go to your Dashboard and find your Hermes Agent instance.
2. Click the Terminal button to open the web terminal.

Step 2: Run the Auth Command

In the terminal, run:

hermes auth add openai-codex --type oauth

The terminal will print a device code (e.g. V0SZ-CL1YM) and a URL. Open the URL in your local browser — on your computer, not inside the terminal.

Step 3: Approve the OAuth Request

1. Visit https://auth.openai.com/codex/device in your browser.
2. Enter the device code shown in the terminal (e.g. V0SZ-CL1YM).
3. Sign in with the OpenAI account that has your ChatGPT Plus subscription and approve.
4. Once approved, the terminal will confirm: Added openai-codex OAuth credential #1: "openai-codex-oauth-1". The credential is stored at /opt/data/auth.json inside your container.

Step 4: Switch the Model

Send this command from any connected chat surface — Telegram, Discord, WhatsApp, or the web chat in your Dashboard:

/model openai-codex/gpt-5.4 --global

That sets model.default = openai-codex/gpt-5.4 and inference.provider = openai-codex in your container's config. The change takes effect on the next message — no restart needed.

The Hermes Dashboard model picker doesn't yet expose Codex as a one-click option, so the chat command is the supported path today. We'll surface it in the picker in a future update.

Step 5: Verify It Works

Send a message to your bot on Telegram, Discord, or whichever platform you connected. The response should come from GPT-5.4 via your subscription. You can also check the active model and credential status with:

hermes auth list

Available Models

Important Notes

• Weekly quota limits: ChatGPT Plus has a weekly usage cap for Codex access. Heavy usage may hit limits before the week resets.
• Token refresh: OpenAI uses rotating refresh tokens. Occasionally you may need to re-run the auth command if the session expires.
• For production workloads: If you need guaranteed uptime and no quota limits, consider using an OpenAI API key instead.

Other Ways to Bring Your Own Key

Don't have ChatGPT Plus? You can still bring your own API key from any provider:

• OpenAI API — Pay-per-token, no subscription needed
• Anthropic Claude — Claude Sonnet 4.6, Opus 4.6
• OpenRouter — Access 200+ models with one key
• DeepSeek — Low-cost coding and reasoning

Or skip API keys entirely — OpenClaw Launch includes AI credits and free models out of the box.

Troubleshooting

Auth command shows no device code

Make sure you're running hermes auth add openai-codex --type oauth (not the OpenClaw command). The terminal must be the Hermes web terminal, accessible from the Dashboard.

Token refresh errors

If your credential stops working, re-run:

hermes auth add openai-codex --type oauth

This replaces the stored credential at /opt/data/auth.json.

Quota exceeded

ChatGPT Plus has weekly limits. If you hit the cap, your bot will return errors until the quota resets. Upgrade to ChatGPT Pro ($200/mo) for higher limits, or switch to an API key for unlimited usage.