Hermes Agent on your ChatGPT plan: setup, login recovery and billing checks

Hermes Agent documents a ChatGPT subscription route: run hermes model, select ChatGPT or Codex Subscription, complete the login and choose an available model. Check that provider before starting work: a direct OpenAI API key or a Nous Portal subscription is a different billing route.

This guide was checked on 26 October 2026 against the current maker repository. It covers connection and troubleshooting, not a claim that every ChatGPT plan gets identical models or unlimited agent runs. Start from the Hermes listing or compare other apps in the ChatGPT hub.

Which Hermes provider uses your ChatGPT subscription?

The maker’s provider guide names ChatGPT or Codex Subscription and identifies its provider as openai-codex. It separately lists OpenAI API access and Nous Portal. Choose by the account route, not merely by seeing a GPT model name.

Route	What the maker documents	What to verify
ChatGPT or Codex Subscription	ChatGPT OAuth through the Codex provider	Authorized account and available models
OpenAI API direct	OpenAI API-key provider	Separate API billing and any spending limits
Nous Portal	Nous subscription gateway	Which Nous subscription or connection funds inference
Claude Subscription DirectSDK	Experimental plugin driving official Claude Code	Plugin requirements and the CLI’s Claude account

The current repository’s Nous Portal page describes inference billed against a Nous subscription. Do not read “GPT available” as “my ChatGPT allowance pays for it.” If your installed Portal flow offers additional account-linking choices, inspect that specific consent and billing screen before using it; this guide does not assume its behavior.

How do you connect an existing Hermes installation?

Run hermes model and select the subscription provider. The maker documents device-code authorization by default: open the supplied URL and complete authorization on the provider’s site. Enter the code only in that sign-in flow, not into a conversation or support ticket.

Choose a model offered to the account and try a small, read-only task before enabling scheduled work or write-capable tools. Our suggested check is to ask for a summary of one non-sensitive document, then inspect the selected provider and account usage. This is a setup recommendation, not a hands-on performance result.

Hermes says this direct provider does not require installing Codex CLI. It manages its own authentication state and can also import an existing CLI login. Never send those authentication files to someone troubleshooting your setup.

For the native Codex route instead, OpenAI’s authentication guide explains ChatGPT sign-in versus API keys. The subscription setup guide explains that distinction across apps.

What should you do when Hermes login stops working?

First distinguish an authorization failure from an exhausted usage allowance. The maker’s provider guide describes terminal refresh failures, including revoked grants and invalid_grant, as reasons Hermes stops replaying the failed credential and requests reauthentication.

Its documented recovery is hermes auth add openai-codex, or reopening hermes model and selecting ChatGPT or Codex Subscription. Complete one fresh provider login rather than repeatedly rerunning the failing task.

For organizations that disable device-code authorization, the guide documents hermes auth add openai-codex --browser. That browser flow uses a local callback, so an SSH or remote-host setup needs its own networking arrangement. Follow the maker’s OAuth-over-SSH instructions rather than exposing a login service publicly.

Before reporting a bug, record the Hermes version, provider identifier, model, error category and whether the problem occurs before or after authorization. Remove credentials and private task content from logs. A successful login alone is not proof of remaining allowance.

Does your plan cover every tool and background request?

Do not assume it. Hermes’s provider guide says auxiliary tasks such as summarization and vision can use separately configured models; its default automatic route follows the main chat model, but individual overrides are possible. Inspect auxiliary and fallback settings as well as the main provider.

OpenAI’s pricing documentation distinguishes subscription usage from API token pricing. This guide deliberately gives no fixed prompts-per-day estimate. Check current entitlement and remaining usage for the selected model instead of relying on a community token count or on what another app permits.

Before adding unattended jobs, decide whether failure should stop the job or switch to a separately paid route. A fallback is a billing decision as well as an availability feature. Start supervised and require approval before sending messages, changing production data or spending money.

Can Hermes use your Claude subscription instead?

Nous publishes Claude Subscription DirectSDK (Experimental). Its README describes an unmodified official Claude Code executable, requires Hermes 0.21.4 or newer and explicitly limits its readiness claims. It is not the same selection as the built-in Anthropic provider.

Keep this route separate when troubleshooting or comparing costs. Review the maker’s requirements, account extra-usage settings and auxiliary routing before trying it. Our Claude subscription guide explains the broader boundary; the Claude hub lists other compatible apps.

For a different general-purpose agent on ChatGPT, see OpenClaw’s setup and billing routes. Do not transfer its model defaults or login instructions to Hermes without checking Hermes’s own documentation.