32dots HEIDELBERG AI
Curriculum › Hermes course › hermes-00
Session 0 easy

Install Hermes & connect a model

USE 0 - 25 min

Get Hermes running — and decide where its brain lives

Unlike the n8n course, there is no shared server to log into. Hermes is yours: it runs on your machine and connects to a model you choose. That is the whole point of a personal assistant — your data, your agent. You install it once — a desktop app on macOS or Windows, or a one-line terminal install on Linux (or any OS) — then pick one of two paths for the model: (A) a cloud provider, or (B) a local model served by LM Studio on your own machine or network.

  1. 1 Easiest — the desktop app (macOS / Windows). Go to https://hermes-agent.nousresearch.com/desktop and download the Hermes Desktop installer: a .dmg for macOS 12+, an .exe for Windows 10/11. Run it — it installs both the desktop app and the hermes command-line tool.
  2. 2 Linux, or the terminal on any OS. Install via script: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash, then source ~/.bashrc. (Windows without the GUI: in PowerShell run iex (irm https://hermes-agent.nousresearch.com/install.ps1).)
  3. 3 Run the setup wizard — open the desktop app and follow its first-run screen, or in a terminal run hermes setup. Either way it asks for a provider and model and writes ~/.hermes/config.yaml.
  4. 4 Path A — cloud model: choose a provider (openai, anthropic, gemini, openrouter…) and paste its API key. Easiest, costs per token. (Or hermes setup --portal for a managed Nous Portal model.)
  5. 5 Path B — local model (LM Studio): in LM Studio, load a model and start its server (Developer tab → Start Server, default port 1234). Then in ~/.hermes/config.yaml set provider: custom and base_url: "http://localhost:1234/v1" (or a LAN IP like http://100.79.134.37:1234/v1 for a model served on another machine). Put any value for OPENAI_API_KEY in ~/.hermes/.env — local servers ignore it. For a full step-by-step walkthrough, see the LM Studio course at /course/lmstudio.
  6. 6 Start it: open the Hermes Desktop app, or run hermes --tui in a terminal. Type a first message: Summarise what you can do in 5 bullets.
  7. 7 Confirm persistence: end the session, then run hermes --continue — it should reload where you left off. If anything is off, run hermes doctor.
✓

Hermes replies from the model you chose, and `hermes --continue` reloads your session. You now own a running agent.

UNDERSTAND 25 - 45 min

What you just stood up — and the one rule that bites

You did not install an app; you installed an agent and pointed it at a brain. Before the parts, here is the map — the Hermes Desktop window you are now looking at. The left sidebar is how you move between the agent's surfaces; the bottom status bar tells you it is alive.

Hermes Desktop at a glance — here running on a local Qwen3.6 model (bottom-right), i.e. the local-model path from this lesson. You will meet each labelled surface in a later lesson.
  • New session — start a fresh conversation/run with the agent.
  • Skills & Tools — the two things the agent acts with: skills (procedures it has learned and saved, see lesson 5) and tools (built-ins like terminal, web search, and file access it can call).
  • Messaging — connect chat apps (Telegram, Signal, Slack…) so you can reach the agent from your phone (lesson 3).
  • Artifacts — everything the agent produced or touched this session: files it wrote, links it opened, images. This is your record of what it actually did — the panel you check to verify its work (lesson 7).
  • Status bar — Gateway (are your chat channels connected?), Agents (running agents / sub-agents), Cron (scheduled standing errands, lesson 6), and the model it is running on.
Key concept

Hermes is four separable things: (1) a process running on your machine, (2) a model you point it at and can swap any time, (3) memory that persists across sessions, and (4) tools and skills it can use. The model is the brain but not the assistant — you can change it without losing your memory or skills. The rule that bites: the model must support a large context (≥64k tokens), and a tiny local model will limit what the agent can actually do. Cloud = more power, pay per token, data leaves your machine. Local = free and private, but only as capable as the model your hardware can run.

  1. ?If you swap the model in config.yaml from a cloud model to a local one tomorrow, what do you keep and what changes?
  2. ?Why does an always-on agent with tools and memory deserve more caution than a chat window you open and close?
  3. ?For your work, name one task you would only ever run on a local model — and one where the cloud is worth the cost and the data exposure.
  4. ?What would break if you connected a model with only 8k context?
BUILD 45 - 60 min

Make the model choice deliberate

Most people accept the default and never think about it again. You will make the choice on purpose — because for a private assistant, where the brain runs is a real decision.

Your task

Get a working reply on your chosen path, then prove you understand the switch by inspecting (or trying) the other one. Decide your default and write down why.

  1. 1 Confirm your primary path works end to end with a real question from your field.
  2. 2 Run hermes model and look at the providers offered — locate where you would switch between cloud and custom (local).
  3. 3 If you chose cloud, open config.yaml and write down exactly what two lines you would change to point at a local LM Studio server. If you chose local, note which cloud provider you would fall back to and why.
  4. 4 Run hermes doctor and confirm a clean bill of health.
  5. 5 Write one sentence: which model is your default, and is your reason privacy, cost, or capability?
Deliverable

A screenshot of a working reply, the model path you chose, and one sentence justifying it (privacy / cost / capability).

Self-check · tick before you mark done

✎A personal assistant only earns its keep if you keep it running. Where will yours live — your laptop, a spare machine, a home server — so it is there when you need it tomorrow morning?