Hermes

Lo configure con OpenRouter y nvidia/nemotron-3-super-120b-a12b:free

Hermes Agent es bastante sencillo de usar. Aquí va una guía rápida:

Instalación

Un solo comando instala todo automáticamente:

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

Instala Python, Node.js y todo lo necesario. No requiere sudo.

Configuración inicial

hermes setup

El asistente te guía paso a paso: elegís el proveedor (ej. Google Gemini OAuth), autenticás, y listo.

Comandos básicos

hermes          # Abre el chat interactivo
hermes model    # Cambiás el modelo/proveedor
hermes tools    # Configurás qué herramientas tiene acceso
hermes gateway  # Conectás a Telegram, Discord, WhatsApp, etc.
hermes update   # Actualizás a la última versión

Uso diario

Una vez dentro del chat, simplemente escribís lo que necesitás en lenguaje natural, por ejemplo:

  • “Buscá información sobre X y hacé un resumen”
  • “Revisá este código y encontrá el bug”
  • “Recordá que mi proyecto está en ~/proyectos/miapp”

Hermes recuerda conversaciones anteriores y va aprendiendo tus preferencias con el tiempo.

Si venís de OpenClaw

hermes claw migrate

Importa automáticamente tus configuraciones, memorias y skills.

Yes, Hermes Agent works great with free models! Here’s the breakdown:

Truly free (local via Ollama):

The best completely free model for Hermes Agent is Qwen3.5 27B running locally through Ollama — no rate limits, no API keys, no usage caps. One important caveat: Hermes requires a 64K minimum context window, which rules out many smaller 7B models. For 8GB VRAM setups, Qwen3 8B has the most reliable tool-calling in its class.

Free cloud tiers (no hardware needed):

  • OpenRouter — sign up for a free account, grab an API key, and pick a free model like nvidia/nemotron-3-super-120b-a12b:free. This is the easiest starting point for beginners.
  • Google AI Studio (Gemini 2.5 Flash) — free tier with 15 requests/minute and 1,500 requests/day, which is enough for light agent use. Note that Pro models are now paywalled.
  • Groq (Llama 4 Scout) — very fast inference, with a free tier of roughly 14,400 requests/day on smaller models, good for testing before committing to paid usage.

Worth noting: Hermes was specifically designed by Nous Research to work well with small and free models — it squeezes more performance out of them than alternatives like OpenClaw would. So you’re in good hands using it with a free model.

Quickest path: sign up at openrouter.ai, get a free API key, and pick a :free model when setting up Hermes.