Hermes Agent Workshop

Before the workshop · Optional

Pre-work Setup ​

This is optional, but recommended. If you can get Hermes installed and a model connected before you arrive, you will have more time to build your own agent during the workshop.

Time
20-40 minutes
Status
Optional, but worth it.

Official docs

Use the official Hermes docs if anything here has changed.

1. Install Hermes ​

Official install guide: Hermes Installation Guide

Default Linux, macOS, and WSL2 install path:

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

Windows (native)

ps1
iex (irm https://hermes-agent.nousresearch.com/install.ps1)

2. Configure Your Model ​

Start by making sure you have one working model path ready to power your agent.

True-free path if you do not already have a subscription: NVIDIA Build ​

If you need a no-card path, create an NVIDIA Build API key: https://build.nvidia.com/models.

Use it with Hermes as an OpenAI-compatible custom endpoint:

text
Base URL: https://integrate.api.nvidia.com/v1
Model: nvidia/nemotron-3-ultra-550b-a55b

NVIDIA Build is the recommended true-free workshop path. Its free limits are not guaranteed, but it is usually a better no-card fit than OpenRouter's true-free tier for live experimentation.

Most reliable free-model path: OpenRouter with $10 credit ​

If you want the fewest surprises on free models, put $10 of credit on OpenRouter: https://openrouter.ai/.

That $10 is not about buying paid inference for the workshop. It unlocks OpenRouter's bumped free-model rate limit. Free models (:free) are capped by lifetime credits purchased:

  • Under $10 purchased: 50 free-model requests per day
  • $10+ purchased (lifetime): 1,000 free-model requests per day

A single complex agent task can burn 5-20+ model requests, so the no-credit free tier is only a smoke test. The $10 unlock is what makes OpenRouter usable for a 4-hour workshop while still staying on free models. Free-model usage does not spend those credits.

Subscriptions You May Already Have that Work With Hermes ​

  • ChatGPT ($20 and up plans): choose OpenAI Codex. Uses ChatGPT/Codex OAuth. Good workshop path if your account has Codex.
  • GitHub Copilot paid plans: choose GitHub Copilot. Uses an OAuth/device-code flow.
  • Claude Max plans ($100 and up plans): choose Anthropic.
  • Google/Gemini accounts: choose Google Gemini OAuth. Browser PKCE login; docs note free-tier support. Gemini API keys also work.
  • Grok paid plans: choose xAI Grok OAuth.

Run the model picker:

bash
hermes model

3. Test the CLI ​

Start Hermes locally:

bash
hermes --tui

Ask something simple, then inspect the available tools and skills:

bash
hermes tools list --platform cli
hermes skills list

If those commands work, you are ready for the main workshop.

4. Optional advanced setup: Docker Backend (10-20 min) ​

The default local terminal backend is the lowest-friction path.

If you would rather not give Hermes direct access to your file system, set up the Docker backend ahead of time so commands run in a container instead of on your host.

Docs: Docker backend

This is optional. Skip it unless you specifically want more isolation.

5. Optional: Telegram Gateway (5 minutes) ​

Telegram is a good first gateway because the setup is simple: about 5 minutes from new account to connected. Install the desktop app, create an account, then run:

bash
hermes gateway setup

Docs:

Built by human and Hermes Agent collaboration.