You've read about agentic AI. You've seen what it can do. Now it's time to stop talking and start building.
There's only one way to truly learn agentic AI: build with it. Reading about it, watching videos, thinking about it — none of that teaches you how it actually works, or what it can really do. You learn it through the process of building something real with it, one iteration at a time.
And that means you need a live agent to work with. Not a chatbot you talk to. A real agent — one that remembers, works on long multi-part projects, and uses tools to interact with the world. That's the difference between reading about the windmills and actually tilting at them yourself.
So this is where we move past the talking and get to the doing. On this page we'll help you get your own working system up and running. It's simpler than you think — you don't need to be a programmer, and you don't need a monster computer. You need a little patience, a willingness to be a beginner for a short while, and a live agent to work with.
This section covers the essentials: what you need to get started, the one big decision you'll make about your agent's brain, and how to get a working agent up and running. For the finer points — costs and the payoff, choosing a model, and running locally — use the links below.
The good news is that the bar for getting started is genuinely low. Here's what you need:
That's really it. No special hardware, no programming experience, no engineering degree. If you have a computer that can browse the web and run ordinary software, you have enough to get started.
This is the one thing most newcomers misunderstand, so let's be clear about it from the start: an agent is not a model. Hermes is the framework — the system that gives the agent its memory, its tools, its ability to work on long projects and talk to the real world. But Hermes does not, by itself, provide the intelligence. That intelligence comes from a large language model (LLM) that Hermes connects to.
Think of it this way: Hermes is the body, and the LLM is the brain. Hermes provides the structure — the memory, the skills, the tools, the hands. The LLM provides the thinking — the understanding and reasoning that turns your requests into real actions. The agent is the two working together as one system.
So before you can do anything, you need to decide where that "brain" will live. That's the one big decision, and it comes down to two options: local or cloud.
These are the parts of the story we've split out so each one gets the room it deserves. Read them in any order — they're companions to this getting-started page.
How tokens work, pay-as-you-go vs flat fees, and the big one — build once, run free forever.
Read more →Where your agent's brain lives: local vs cloud, and how to choose without burning tokens.
Read more →Run a model entirely on your own machine — full privacy, no ongoing cost, and the GPU question.
Read more →This walkthrough is written for Windows, because it's what most of our visitors use — it remains the operating system of choice for home desktops and small business. But the concepts here are the same on any system. We'll focus on Windows to get you going, and if there's enough interest, we may add guides for other operating systems later. For now, the important thing is that you get a live agent to work with.
You've already made your big decision from earlier: cloud or local. For your first time, we recommended cloud. So this walkthrough assumes you're using a cloud model — the easiest path to a working agent. The whole process takes about ten minutes.
Go to the Hermes website and download the Hermes Desktop installer for Windows — it's a normal .exe file. Then double-click it to run it, just like any other Windows program. The installer handles everything for you — Python, the required tools, and Hermes itself — and it sets up a desktop icon you can use to launch the app.
Prefer the command line? If you're comfortable in a terminal, you can install the same thing in one step by opening PowerShell and running:
Both methods give you the same working Hermes installation — they share the same data folder, so you can switch freely between them. The desktop installer is the friendliest way to start; the command line is a great option if you're already at home in a terminal.
With the desktop installer, the easiest way in is to double-click the Hermes icon on your desktop — it opens the app window, just like any modern program. (You can also open the app any time by typing hermes desktop in a terminal.) The first time it launches, it may finish setting up — give it a moment.
Hermes needs a "brain" — that's the model you chose earlier, and the API key you set up. You have two honest paths here:
Path A — Bring your own provider (more flexible). If you want the freedom to choose your own providers and models, open a terminal and run:
This walks you through picking a provider (OpenRouter, OpenAI, Google, Anthropic, and many others) and a model. When it asks for an API key, paste the one you created ahead of time. This path gives you the most control over which models you use and how you pay for them.
Path B — One simple subscription (easiest). If you'd rather not think about providers and keys at all, you can run:
That sets up a Nous Portal account — one subscription covering hundreds of models — and configures Hermes to use it in a single step. It's the fastest way to get going and matches the flat-fee idea we talked about earlier.
There's no wrong choice. Path B is the quickest and simplest — ideal if you just want to start building today. Path A trades a few extra minutes for more flexibility and control over your models and costs. Many people start with one and try the other later.
Back in your app window, you can start chatting right away. Ask it something simple to confirm it's alive — then get ready to be surprised at how quickly the real work begins.
A quick note on the "hermes" command: typing hermes in a terminal opens the text-based interface (the CLI), not the app window. Both connect to the same agent — the app is one way to reach it, and the terminal is another. We'll cover all the ways to talk to your agent in the next page.
Don't let it sit idle. Give it a small, real job — something you actually want done. A good starter: "Write a short list of things I should know about getting started with agentic AI, and save it as a file on my desktop."
Watch what happens. It will use its tools, create the file, and tell you it's done. Then check the file — and if it's not quite right, tell your agent what to change. That build-and-improve loop, right there, is the whole skill. You're learning agentic AI by doing it, exactly as you should.
And that's the entree — your very first working agent, up and running in about ten minutes. Now the real learning begins. Go build something.
