Master Hermes AI Agent Setup — Step-by-Step Guide & Transcript

[00:01] ahead of you on AI agents, they're really not because this is so new that hardly anyone has figured out how to build one properly yet. So, in the next completely set up with Hermes. And by the end, you will be ahead of nearly

[00:14] sidelines. If you're not fully sure what Hermes even is, it's an AI agent, which means instead of answering one question at a time, like ChatGpt or Claude, it runs on its own schedule and handles background tasks automatically. You can

[00:29] have it send you a morning summary, pull some specific stats for you every week, hand you back the finished result, all without you being at your computer. And there is a reason this is the one worth

[00:42] completely free, and the team behind it is a well-known AI lab called Nouse Research. The memory, the scheduling, the way it learns your work over time, all of that is already built into it. So, none of it is something you have to

[00:55] set up yourself. You're also not tied to one company's AI because it plugs into service called Open Router and you can switch whichever model it uses whenever you want. And this isn't a tiny side project either since more than 181,000

[01:09] developers are already backing it on GitHub, which is a strong signal for though, and it's the whole reason you can't just run an agent off your laptop. An agent only does anything while it's actually running, and the second you

[01:22] close your laptop, it goes to sleep. So, an agent kept on your own computer would stop the moment you step away. That's why it needs to run somewhere that never VPS, which is really just a small

[01:34] the time, so your agent keeps running, like the scary technical part, but it barely takes any effort at all. Hostinger is the tool behind all of that. So, let's get you set up. The link

[01:47] onto the deploy page with the discount already on it, so you don't need to enter a code. The plan you want is KVM2, and it's already the one selected with more than enough room to run Hermes, so you can leave it exactly as it is. Next,

[02:01] it asks how long you want to pay for, and it starts on 24 months, but you're free to bring that down to 12 or even a single month. 12 is the one I'd pick still there at that length. Right under that, you can flip on daily backups and

[02:14] choose your server location and just pick the one closest to you. Look over the summary on the right, then click continue, pay, and make your account. setup screen, and it's worth slowing down here for a moment. It hands you a

[02:27] username and a password. Leave the username as a Hermes, but that password is generated just for you. So, copy it and keep it somewhere safe because logging into your server later depends on it. After you hit deploy, Hostinger

[02:39] goes off and builds a server and installs Hermes for you. And a couple of minutes later, your dashboard loads up. On there, Hermes shows up under a part called Docker Manager. You'll also see something called Trafic right below it.

[02:51] and Trey only handles the networking so you can completely ignore it. Your Hermes project will have four random characters on the end of its name and that in mind. Up in the top right there's a button labeled terminal and

[03:04] clicking it pops open a window called the web terminal. What's coming up is the only part that looks technical but it's far simpler than it seems. So stick with me. The web terminal is really just a window where you type a few commands

[03:16] it this one time during setup. I'll start by opening the Hermes folder. So I'll paste in this command with those four characters from the dashboard added to the end and press enter. Then I'll paste the next command to open Hermes

[03:29] itself. Press enter. And now I'm talking to it directly. After that I'll type Hermes setup. Hit enter and the setup wizard takes over from there. It starts by asking if you want the quick setup or the full one. And full is what I'll

[03:41] choose so we can get to it with more detail. Its first question is which AI provider you want. And I'm going with open router. The reason is that one open router key opens up more than 400 different AI models and you can jump

[03:53] command. So nothing you choose here locks you in. To grab the key, I'll head to open router. I already have an account, so it will take me straight to the API key section. And I'll hit new key, which is basically a password that

[04:06] you. While you're there, I'd put a spending cap on it somewhere around $25 over what you meant to spend while you experiment. You'll also need to drop a little credit onto your open router account because even after the key

[04:21] exists, it won't actually do anything until there's a balance behind it. With into Hermes and choose the model it runs on. If keeping costs near zero is the goal. Open router has free models and one called Owl Alpha is built for agent

[04:35] work and won't charge you a scent. If you'd rather have stronger answers, something like GPT 5.5 or Claude, both work great for that, and most tasks land couple of dollars, depending on their size. There's also a setting called auto

[04:49] routing, where Hermes decides the model for each task by itself, handing the easy jobs to a cheap model and the harder ones to a stronger model, so you save money without ever thinking about it. Counting from the moment you hit

[05:01] deploy to Hermes actually being ready to talk, it's roughly 10 minutes. And the server to finish. Hermes is running right now, but the only way to reach it is through that terminal window. And keeping that open forever would defeat

[05:14] the whole point of an agent that runs on its own. And sorting that out is where straight from your phone. The wizard carries on from where we stopped. And you want to connect. The list has Telegram, Discord, Slack, WhatsApp,

[05:29] Telegram because it is free and quick and runs on basically any device. It needs two things from you. A bot token and your Telegram user ID. And grabbing both takes about a minute. For the token, I'll open Telegram, look up a

[05:43] user called botfather, and send the message/ newbot. Botfather has me name the bot and give it a username ending in the word bot. And once that's done, it hands back a long token, which I copy and paste into Hermes. For the user ID,

[05:55] I'll look up another bot called userinfobot. Send/start and it shows my number right away, which I also paste into Hermes. Hermes then asks if that should be my home channel, which is simply where it sends scheduled results

[06:07] and notifications. And I'll press Y to lock it in. That wraps up the wizard, but there's a single step left before your agent is actually live. You need to piece that keeps Hermes connected to Telegram and listening for whatever you

[06:20] this. Anytime you want to start your agent quickly from the Docker page, just click the terminal next to your Hermes container and paste the getaway command in the terminal. Now I'll open my bot in Telegram, send a message, and there it

[06:33] you've officially built and deployed your first AI agent. If the gateway ever stops, your messages get no reply, and your scheduled jobs won't fire. So that's the first thing to check anytime your agent stops responding. One thing

[06:46] I'd set up right away is a Telegram group for your bot with topics switched keep separate threads for different for your numbers, and one for personal

[06:58] stuff, so everything doesn't end up in one messy chat. For topics to work, head settings, and it will take you to botfather and switch group privacy off. Otherwise, it can't read the messages inside the group. You'll also want to

[07:11] make the bot an admin of that group, so it's actually allowed to reply. And if talking is easier than typing, Telegram lets you send voice messages and Hermes turns them into text and answers. So you can hand it a task even when you're away

[07:23] from your desk. With your agent live and listening, you can start teaching it to that's you finished with the terminal for good. From here on, everything, every task, every schedule, every preference, all happens inside that same

[07:36] scheduling since it's the feature that does the most for you on its own. Setting up a scheduled task is as simple as telling Hermes what you want in plain English. If I type, "Send me a tech news summary every morning at 8:00 a.m.,

[07:49] where to send it, and reads the whole thing back to me to confirm. I'll add one more and type." Every Monday at 8:30 a.m., pull the latest sales data, and send me a summary of how this week did compared to last week. Hermes makes the

[08:02] job, shows me the timing, and from then on, that report turns up in my Telegram every Monday morning without me typing a single thing. Anytime I want to see what's active, I type, "Show me a list of all current jobs and it lays them out

[08:14] with their times and I can pause, tweak, or delete any of them right there in the to underestimate because once a handful of these are running, your agent is single day while your attention is somewhere else. Skills run on that same

[08:29] hands-off idea. Each time you finish something that took a bit of back and forth, Hermes studies what actually worked and saves the whole thing as a spent 15 minutes shaping a social media post just right, fixing the tone and the

[08:43] that, and next time you simply say, "Create a social media post," and it runs the saved process and nails it in one shot. You can also make a skill on purpose by saying something like, "Take what we just did and save it as a skill

[08:56] called niche hookwriter. And from then on, you call it up by name." The last piece is memory, and it runs on three simple files you will never need to open agent's personality, whether you want it relaxed and friendly or short and to the

[09:09] point. The second sets the rules for how it does your work, things like the writing tone you like or how you want things laid out. And the third is its job, your projects, and your preferences. And it keeps that updated

[09:22] on its own as the two of you talk. You never edit any of these by hand. You just tell Hermes in Telegram. If I type, remember that I prefer short and direct answers with no long explanations unless I ask. It saves that and uses it in

[09:35] you tell it about how you work, the sharper every answer gets because it uses your personality, your rules, and your context before sending anything back. And once your tasks are running and your skills are building on their

[09:48] knowing before you hand it anything bigger. Keeping it safe and keeping it cheap. Hermes is locked down by default, and there are four built-in protections worth knowing, especially early on. The first is that only the people you

[10:01] approved during setup can message your agent. So even if a stranger stumbled onto your bot in Telegram, it would just ignore them. The second is that Hermes runs inside its own isolated space on the server. So even if something broke,

[10:13] it couldn't touch anything else on your system, and that comes free with the one-click install. The third is that anytime your agent goes to do something risky, like deleting files or making a big change, it pauses and asks for your

[10:25] okay first. So read those messages before you approve them. And the fourth is that every new skill you add gets scanned for unsafe code before Hermes will run it. For your first week, my honest advice is to keep it small, stay

[10:37] on Telegram, pick one or two tasks that solve an actual problem you have, and leave anything sensitive disconnected. Things like email or banking until you trust how it behaves. The other thing to keep an eye on is cost, and you keep it

[10:50] down by matching the model to the job. There's no reason to have your priciest model answering simple questions, so you leave the cheap or free ones on the models for the real thinking. Inside Telegram, typing /model shows what

[11:04] you're on and lets you switch to any other model on the spot. And if you want smarter, you just say, "Run the daily news summary on Opus 4.8 instead of my default." And Hermes moves that single job over while everything else stays

[11:18] cheap. Hermes itself is completely free and open- source. It comes with over a 100 ready-made skills, works with 18 messaging apps, and connects to more than 200 AI models. So, the only things you ever pay for are the server it runs

[11:31] on and the small cost of the AI it uses. Hostinger took the hardest part of all this, setting up the server and turning it into a single click. And that is the go click the link below and get your own

[11:44] never really behind. You were just one setup away and now that's done. Thank you for watching and I'll see you in the next one.