One Claude Pro subscription, every tool: using `hermes proxy` with Aider, Cline, and Codex CLI

v0.14.0 shipped a one-line command that quietly absorbed a whole class of integration questions the community had been asking for months. Run hermes proxy and you get a local http://localhost:port endpoint that speaks the OpenAI API, backed by whichever OAuth provider you're signed into through Hermes — Claude Pro, ChatGPT Pro, or SuperGrok.

The practical consequence: tools that only know how to speak the OpenAI API — Aider, Cline, Codex CLI, Continue, and the long tail of editor plugins — can now talk to Claude Pro, ChatGPT Pro, and SuperGrok through your existing subscription, without you ever generating an API key.

This is the part that matters: most of these tools require billing relationships per provider. Aider with Claude → Anthropic console account with a payment method. Cline with GPT-5 → OpenAI billing account. Codex CLI → likewise. hermes proxy collapses all of that into one localhost endpoint that authenticates with the OAuth session you already have.

Here's the exact workflow.

0. Prerequisites

• •Hermes Agent v0.14.0 or newer installed and configured (hermes --version to check)
• •At least one OAuth subscription linked through hermes model — Claude Pro, ChatGPT Pro, or SuperGrok

If you've never logged into an OAuth provider, the first time you run hermes model it walks you through the OAuth flow per provider. Pick whichever subscription you actually pay for.

1. Start the proxy

hermes proxy

By default it binds to a free port on localhost. The startup log shows the URL — copy it. To pin the port:

hermes proxy --port 11500

Run it as a long-running process — keep it alive while downstream tools are using it. systemd, launchd, tmux, or just a terminal tab all work.

2. Point a downstream tool at the proxy

Aider

export OPENAI_API_BASE=http://localhost:11500/v1
export OPENAI_API_KEY=hermes-proxy  # any non-empty string works
aider --model claude-sonnet-4-6

Aider's --model flag references the model name as the proxy resolves it from your Hermes provider config. If you're signed in via Claude Pro, your default Claude model is available; same for ChatGPT Pro and SuperGrok.

Cline (VS Code)

In Cline's settings, choose OpenAI Compatible as the provider. Set:

• •Base URL: http://localhost:11500/v1
• •API Key: hermes-proxy (any non-empty value)
• •Model ID: the model name from your Hermes config (e.g. claude-sonnet-4-6, gpt-5.1)

Save. Cline now routes its requests through Hermes; your subscription auth handles the rest.

Codex CLI

Codex CLI typically expects OPENAI_API_BASE and OPENAI_API_KEY. Same shape as Aider:

export OPENAI_API_BASE=http://localhost:11500/v1
export OPENAI_API_KEY=hermes-proxy
codex --model gpt-5-codex

If you're signed into ChatGPT Pro through Hermes, the gpt-5-codex model surfaces through the proxy and Codex CLI gets full access without you ever creating an OpenAI billing relationship.

Continue (JetBrains / VS Code)

Edit your Continue config (~/.continue/config.json or the UI):

{
  "models": [{
    "title": "Claude via Hermes",
    "provider": "openai",
    "model": "claude-sonnet-4-6",
    "apiBase": "http://localhost:11500/v1",
    "apiKey": "hermes-proxy"
  }]
}

Same pattern. The "provider" stays openai because, as far as Continue is concerned, it's hitting an OpenAI endpoint.

3. Verify it works

A quick curl test:

curl http://localhost:11500/v1/models -H "Authorization: Bearer hermes-proxy"

You should get back the list of models your Hermes provider config exposes. If you get a 401 or empty list, the proxy is running but Hermes can't see an active OAuth session — re-run hermes model and pick a provider you have a live subscription for.

4. Troubleshooting

• •"Connection refused" — the proxy isn't running. Start it.
• •"401 unauthorized" — the model you requested isn't covered by your linked subscription. Run hermes model and check which providers are active.
• •"Model not found" — model name in your downstream tool doesn't match what your Hermes provider exposes. Use the names from hermes model list.
• •Rate-limited — you're hitting the subscription's rate limits (Claude Pro has ~50 messages per 5h on Sonnet; ChatGPT Pro has a separate quota; SuperGrok has its own). The proxy doesn't add a rate limiter; it forwards whatever quota your provider gives you.

Why this matters

The OpenAI API is the lingua franca of the AI tool ecosystem. Almost every editor plugin and code assistant supports OpenAI first and "OpenAI-compatible endpoints" second. The catch is that "compatible endpoint" almost always means "you need an API key, which means you need a billing account, which means your $20/month subscription becomes a pay-as-you-go account on someone's dashboard."

hermes proxy collapses that problem. Your OAuth subscription stays where it lives. The proxy speaks OpenAI on localhost. Aider, Cline, Codex CLI, Continue — they don't know they're talking to Claude or Grok. They just see an OpenAI endpoint that works.

One subscription, every tool, no API key. That's the whole pitch.

Read more

• •v0.14.0 release notes — hermes proxy announcement
• •Hermes Agent docs
• •Hermes Agent on GitHub