Hermes Agent

Run the Hermes Agent from Nous Research on Nebul models through a custom endpoint.

Hermes Agent is Nous Research's open-source personal agent: a terminal, desktop, and gateway assistant that lives on your own infrastructure, talks to Telegram, Discord, Slack, and WhatsApp, and improves its own skills as you use it. It works with any model endpoint you give it, including Nebul.

Prerequisites

A Nebul AI Studio account with API access
An API key for your active project
A model ID from the Model Catalog

Installation

On Linux, macOS, and WSL2:

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
source ~/.bashrc    # or ~/.zshrc
hermes

Windows, Termux, and other install methods are on the installation docs.

Connect it to Nebul

The guided way is the model setup wizard, run from your terminal outside any chat session:

hermes model

Select Custom endpoint (self-hosted / vLLM / etc.).
Enter the API base URL: https://api.inference.nebul.io/v1
Enter your Nebul API key.
Enter the model name: zai-org/GLM-5.3, or any model your project can access.

Or set the same values directly in ~/.hermes/config.yaml:

model:
  default: zai-org/GLM-5.3
  provider: custom
  base_url: https://api.inference.nebul.io/v1
  api_key: sk-your-api-key-here
  # key_env: NEBUL_API_KEY    # read the key from an env var instead

config.yaml is the single source of truth for the model, provider, and base URL, whichever way you set it.

Run it

Start it with hermes, or keep it running as a gateway and talk to it from Telegram or any other channel you have connected. Inside an active chat, /model switches models without a restart. Your conversation history, memory, and skills carry over between providers.

Troubleshooting

404 on requests: base_url must be https://api.inference.nebul.io/v1, with the /v1 suffix.
model not found: the model name must exactly match a catalog ID, including vendor prefix and casing.
Key not picked up: if you use key_env, that variable must be set in the environment of the process that starts Hermes.
hermes doctor prints provider warnings for anything it can't resolve, and it's the first command to run when a model misbehaves.

Data and telemetry

Hermes runs entirely on your own infrastructure. Model traffic goes only to the endpoint you configure. The only telemetry question comes from the first-run setup, which asks whether to share setup metrics; declining creates no telemetry identity and sends nothing.

For how Nebul treats requests, see Privacy & Security.