To connect Claude to Hermes Agent, run hermes model and pick Anthropic, then sign in with an Anthropic API key (pay per token) or a Claude Max login with extra usage credits. You can also reach Claude through OpenRouter. Here is the setup for each route and how to choose. Commands below come from the official Hermes Agent documentation as of November 2026; check it before you copy anything, because the project ships often.
What does "connect Claude to Hermes" mean?
It usually means making Claude the language model Hermes thinks with. Hermes is the agent: memory, skills, messaging, scheduling. Claude is the brain behind it. That is different from Hermes handing coding work to Claude Code, which keeps its own model and runs as a separate worker. This post covers the first. Our Hermes orchestrator guide covers the second.
Many setups use both: Claude as the model Hermes reasons with, and Claude Code as the worker it delegates builds to. Get the model connection working first. Delegation only makes sense once Hermes answers reliably on its own.
How do you connect Claude with an Anthropic API key?
Create a key in the Anthropic console, set it as ANTHROPIC_API_KEY, and start Hermes with the Anthropic provider. Requests bill per token against that key's organization at standard API prices, separate from any Claude subscription. This route works for anyone, needs no particular plan, and is the one we recommend for a first setup.
export ANTHROPIC_API_KEY=your-key
hermes chat --provider anthropic --model claude-sonnet-4-6The model name above is the example the Hermes docs use. Run hermes model to see the Claude models your key can reach and pick one there. To make Claude the default instead of passing flags every time, set it in your Hermes config:
model:
provider: "anthropic"
default: "claude-sonnet-4-6"Keep the key in ~/.hermes/.env or your shell environment, never in a file you might share. The Hermes docs also accept --provider claude and --provider claude-code as shorthand for the same provider.
Can you use a Claude subscription instead of an API key?
Only on Claude Max with extra usage credits. When you pick Anthropic in hermes model and log in with OAuth, Hermes draws on the extra credits you bought on top of Max, not the usage Max includes for Claude Code. The Hermes docs state that Claude Pro cannot use this route. If you already use Claude Code on the same machine, Hermes can read its stored login.
hermes model # pick Anthropic, then sign inThe Hermes plugin catalog also lists an experimental Claude Subscription plugin that routes Hermes through an installed, logged-in Claude Code CLI. It is marked experimental, so treat it as something to test, not the base of a setup you depend on. If you are not on Max, the API key route is the reliable one.
How do you connect Claude through OpenRouter?
Set an OpenRouter key, pick OpenRouter as the provider, and choose a Claude model by its OpenRouter name, such as anthropic/claude-sonnet-4 in the Hermes docs example. OpenRouter is useful when you want one key and one bill across several model makers, or want to switch between Claude and other models without new accounts.
The trade is one more company between you and Anthropic. Hermes connects to Anthropic natively, so you do not need OpenRouter to use Claude. Use it for the convenience, not because it is required.
Which route should you pick?
Pick the API key unless you have a reason not to. It works on any plan, the bill is easy to read, and you can cap spend in the Anthropic console. Pick the Max login if you already pay for Max and bought extra credits. Pick OpenRouter if you want several model makers under one key.
| Route | Who it fits | How it bills |
|---|---|---|
| Anthropic API key | Anyone; the default choice | Per token, standard API prices |
| Claude Max login | Max subscribers with extra usage credits | Draws on the extra credits, not the base Max allowance |
| OpenRouter | People using several model makers | Per token through OpenRouter |
Why isn't Hermes using Claude after setup?
The most common cause is switching inside a running chat. The /model command only moves between providers you have already set up; it cannot add a new one or run a login. Exit the session, run hermes model, finish the Anthropic setup, and start a new session. Then check that your key or login is the one Hermes picked up.
Two other checks. Hermes needs a model with at least 64,000 tokens of context, which current Claude models clear easily, so this only bites when a fallback points at something small. And if you set fallback providers, a failed Claude call can swap models mid-session without warning you, so look at which model answered before you blame Claude for an odd reply.
What should you do after Claude is connected?
Give Hermes one small job and let it run for a week before you add more. Our getting started with Hermes guide walks through setup, memory, skills and Telegram, and what a realistic first week looks like.
When you are ready for Hermes to hand real work to Claude Code, the Hermes orchestrator guide covers delegation briefs, a work board and verification gates, with a free pack. If you want a guided first delegation, the Hermes first handoff kit walks one through end to end. And if the agent is meant to deliver a service you sell, the Claude Code Service OS packages that offer.
An agent is only as good as what it knows about your business. In Brand Architect, Ophelia builds a knowledge base from your website or materials that any AI tool can read. Build your knowledge base →