Hermes

Hermes Agent is Nous Research's self-improving agent runtime — it works with PrimeRouter out of the box. Use the one-click script and pick the Anthropic (Claude) or OpenAI (Codex) channel when prompted.

Two channels — pick one

Channel	Base URL	Example model
Anthropic (Claude)	https://primerouter.ai	claude-opus-4-8
OpenAI (Codex)	https://primerouter.ai/v1	gpt-5.6-sol

The Claude / Anthropic endpoint must not carry /v1; the OpenAI / Codex endpoint must carry /v1.

Recommended: interactive one-click script

It asks for the channel, your API key, and lets you pick a model from the gateway's live list, then writes ~/.hermes/config.yaml:

macOS / Linux / WSL

curl -fsSL https://primerouter.ai/install/hermes.sh | bash

Windows PowerShell

irm https://primerouter.ai/install/hermes.ps1 | iex

💡 The script is config-only — it never installs or reinstalls the Hermes runtime itself. When the hermes command is missing it prints install hints (see below). Grab your API key from Console → Tokens first.

Then start it:

hermes

One-shot sanity check:

hermes chat -q "Hello, introduce yourself"

Fallback: manual install & configuration

Expand when you want to review the commands or the one-click script is unavailable

Step 1 — install the Hermes runtime (either):

pipx install hermes-agent

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

Step 2 — pick a channel and write ~/.hermes/config.yaml

Replace YOUR_API_KEY with your PrimeRouter token:

Anthropic (Claude)

model:
  default: claude-opus-4-8
  provider: primerouter-claude
providers:
  primerouter-claude:
    api_mode: anthropic_messages
    base_url: https://primerouter.ai
    api_key: YOUR_API_KEY
    default_model: claude-opus-4-8
    models:
      - claude-opus-4-8

OpenAI (Codex)

model:
  default: gpt-5.6-sol
  provider: primerouter-codex
providers:
  primerouter-codex:
    api_mode: codex_responses
    base_url: https://primerouter.ai/v1
    api_key: YOUR_API_KEY
    default_model: gpt-5.6-sol
    models:
      - gpt-5.6-sol

The Anthropic (Claude) channel uses the root URL — no /v1; the OpenAI (Codex) channel uses the /v1 URL. See Available models for the full list.

Step 3 — use it

hermes

A running Hermes needs a restart to pick up the new config.