The Only Hermes Agent Tutorial You'll Need in 2026

[00:00:03]Every AI tool you use has the same problem.

[00:00:08]You open a session, you explain your project, you correct your mistakes, you teach it your preferences, and you get something useful out of it. But then you close the tab and tomorrow it's forgotten everything and you have to start from scratch again. But [music] what if the agent just didn't forget?

[00:00:21]What if it remembered your codebase, your conventions, your corrections, [music] and actually got better every time you used it? Today we're talking about Hermes agent and we're setting it up from scratch.

[00:00:36]So, Hermes agent is an open-source AI agent built by Nous Research, the lab behind the Hermes model series. [music] It's not a chatbot, it's not a coding co-pilot tailored to your IDE. It's a persistent agent that lives in your machine, remembers what it learns, writes [music] its own skills, and gets measurably better the longer you use it.

[00:00:53]You just need to think of it like using a junior engineer who never quits, never forgets, and writes their own training manual after every task. It's like a French venture. Since launching in February 2026, it's crossed 157,000 GitHub [music] stars and it's currently the number one framework on OpenRouter for global token usage.

[00:01:09]Now, I'm sure you must be asking, "Why is it gaining so much traction?" And that's because of these four things. So, first, persistent memory. Your preferences, your project context, your corrections, [music] they survive between sessions. You don't have to re-explain yourself like you do with other tools. Second is self-evolving [music] skills. When Hermes solves a hard problem, it writes a reusable skill file. Next time it [music] hits something similar, it loads the skill instead of trying to figure it out from scratch. Third is the messaging gateway. You can talk to it from Telegram, [music] Discord, Slack, WhatsApp, Signal, on 20-plus platforms.

[00:01:42]And so, you start a conversation in Telegram and you pick it up in your terminal. [music] Fourth is cron scheduling. Now, this runs tasks while you sleep, whether it's daily reports, [music] nightly backups, or morning briefings, all unattended.

[00:01:54]And here's the part that matters for today's video. [music] It works with Grok through the Super Grok subscription or XO Auth. So, if you've got a Super Grok sub, you're already set up for a model provider. So, for India, it's like 700 bucks a month and you get unlimited text to Hermes agent. All right. So, here's the plan. We'll install Hermes from zero. [music] We'll walk through everything it ships with. We'll go deep on the skills and memory system, which is the real magic. We set up Telegram so you can talk to it from your phone and we'll also set up cron jobs so it runs 24/7. So, now let's get into it. So, installation for this is super easy.

[00:02:26]It's just one command and this is what the command is. So, this handles everything. It installs UV, Python 3.11, Node.js, Ripgrep, FFmpeg, all of it. No sudo needed.

[00:02:37]Works on [music] Linux, Mac OS, and WSL 2. So, if you're on Windows, you can use WSL 2. Native Windows supports exist, [music] but it's early beta. So, now the setup wizard.

[00:02:47]Okay. So, this setup wizard walks you through everything. Model provider, tools, terminal back end. Now, for the model provider, this is where most people get stuck deciding. And [music] here's what I'd recommend. If you have your Super Grok subscription through X, use the XO Auth option. Honestly, this is the simplest path right now because you don't need a separate API key. Just authenticate it and you're in. If you want access to 200 plus models, all you need to do is go to OpenRouter because with one API key, you [music] get every model. And if you want to use Claude or GPT directly, those work, too. Hermes is model agnostic, so you can swap models anytime with Hermes model. No code changes whatsoever.

[00:03:31]So, [music] once setup completes, you start Hermes with one word. And all that word is >> [music] >> Hermes.

[00:03:38]And that's it. You're in. Full interactive CLI with tools, memory, skills, and streaming [music] output.

[00:03:45]Here's another thing you need to know.

[00:03:46]If you're coming from Open Claude, here's the migration command that you need to use. Hermes Claude [music] migrate.

[00:03:52]Now, when you use this, it imports your settings, your memories, your skills, and API keys automatically. This also detects your open claw install and offers to migrate during the first setup already we're running. So, let's see what we actually have.

[00:04:04]>> [music] >> Okay. Now, Hermes ships with over 70 built-in tools across 28 tool sets and 123 bundled skills before the agent writes one single line of its own. So, let's break down what's actually happening here. So, when we're talking about tools, you have web search, you have [music] full browser automation, shell execution, image generation, vision analysis, text to speech, [music] code execution, all built-in.

[00:04:27]Now, when we're talking about the skills, we have GitHub workflows, [music] we have Obsidian, we have Google Workspace, Linear, Notion, Typefully, Perplexity, Deep Research, plus 100 more. When we're talking about the memory, persistent across [music] sessions. We'll go deep on this in the next section. Now, when we're talking about the profiles, you can run multiple isolated agents, whether it's a programmer, a researcher, designer, each with their own personality, [music] memory, and skills.

[00:04:50]So, you can assign, track, visualize, all from the terminal.

[00:04:54]Now, before we get into the technical details, I want to give you a mental model on how you should think about this. [music] Shawn Wangburg, who built his entire marketing operation on Hermes, has [music] this four-level framework that I think is the cleanest way to think about it. So, level one is one agent, your laptop or a VPS. You talk to it, it learns your style, builds memory, and this is where everyone should start, honestly speaking. Now, when we move on to level two, we're going to be talking about multiple specialist agents, an SEO agent, a dev agent, a personal assistant, each with its own personality and memory, and you can talk to each one directly. Level three is orchestrator plus specialists.

[00:05:27]One front door that routes your work to the right agent. Level four is fully automated team. Cron jobs fire, agents run, workflows unattended, reports land in your Telegram before you even wake up. So, today we're getting you solidly into level one with everything that you need to grow from there. Don't try to jump to level three on day one because that's a trap. Start with one agent, let it learn, then expand. All right, now's the part that I'm most excited about.

[00:05:52]So, now [music] we're actually going to set Hermes up properly, not just install it, but make it yours. I've already sorted my Super Grok subscription and if you haven't just search for Super Grok on X, you get a three-day free trial and then you can decide if you want to keep it. I'm pretty sure you will because it's genuinely worth it. Okay, after the install, just follow whatever the setup wizard asks you to do because once you're through the steps, you'll be inside the Hermes TUI. Now, here's the thing. At this point, you have a working agent. It can read your files, write messages, search the web, use all its tools and skills, but it's still a generic agent. It doesn't know you. You can also hook up additional APIs if you want, but you don't have to. It works fine without them. Okay, here's the best thing that I did for my setup. I've been using ChatGPT and Claude for two to three years now and there's a lot of context about me, my preferences, my thinking style already sitting in these conversations. So, I just went to ChatGPT and Claude and asked them, "Can you write me a soul.md file for me?" An instructional file that captures how I think, how I respond, and what kind of tone I want.

[00:06:51]And then, I took that output and told Hermes to use it. Just drop the soul.md file so the agent knows who I am from the very first message. Now, this is the difference between Hermes being a generic AI wrapper on top of Grok and Hermes being a friend and thinker alongside you. It replies how you want it to and let's not forget that it matches your energy. Now, look at the difference. Before setting this up, you get a standard AI response and after, it actually sounds like someone who gets you. All right, so we've got Hermes running and personalized. Now, let's make it reachable from your phone. The messaging gateway is how Hermes goes from terminal tool to personal agent that lives where you do. One command is all you need and it connects to Telegram, Discord, Slack, WhatsApp, Signal, all 20 plus platforms from a single process. And also, this runs as a system service so it always stays up.

[00:07:33]Now, let me walk through Telegram since that's what most people set up first.

[00:07:37]Now, what you need to do is create a bot through Bot Father on Telegram, get the token, and drop it into your configuration. Then you need to set your home channel. Now, this is the default channel where the agent lives. Start the gateway, send it a message from your phone, and that's it. The agent is now reachable from your pocket. And here's the nice part, conversation continually works across platforms. So, you can start on Telegram, pick it up in your terminal, and then the session state follows you because it's all the same agent underneath. This gateway is what takes Air message from a cool terminal tool to something that you'll actually use every day because you're already checking these messaging apps 20 times a day, and now your agent just lives there. So, now we're moving on to the final piece, which is making Air message work when you're not watching. So, now we're going to talk about cron jobs.

[00:08:19]Cron jobs in Air message are natural language. You don't write cron expressions. You describe what you want in English, and Air message converts it.

[00:08:26]All right, so this is the prompt that I'm going to be putting into Air message, and >> [music] >> honestly, that's it. Now, every morning at 8:00 a.m., a briefing lands in your Telegram before you even had your coffee. Oh, and when we come to deployment, you've got a whole bunch of options. The simplest production setup is a cheap VPS, whether it's Hetzner, DigitalOcean, [music] whatever you prefer. Install Air message, run the gateway as a system service, and it stays up 24/7 even when your laptop is off. It's kind of a painful process if you're doing [music] it for the first time, but if you know how to set up basic cloud infrastructure, and you're fine paying $5 to $10, it's the best thing to do. Now, honest take here, the VPS [music] experience I've had is not as polished as I expected, and I've tried Open Claw as well. And Open Claw was actually a bit better on this front.

[00:09:04]Now, Air message is very, very new.

[00:09:06]They'll [music] get there soon, but right now, the always-on server setup isn't as smooth as the rest of the product. If your laptop is on, Air message does a bang-up job. If your laptop is off, it's running on a Linux server, and you lose access to your local Mac apps and files. And unfortunately, I wasn't able to get that working cleanly. [music] If that update comes, we'll make another video and do this properly, but for now, it's a bit of a duct-tape situation for the VPS part. Keep your laptop on, and it'll be great, [music] but if you want it running headless on a server, just note that it's not there yet. And now it's the question that everyone will be asking, "What about the cost?" Well, the framework is free. You pay for LM API calls, so typically 5 to 10 rupees per task depending on the model. The VPS is 400 to 500 rupees per month and that's your entire AI team running 24/7. Now, this maps to level four in Shawn's framework, the fully automated team.

[00:09:50]Cron jobs fire agents run workflows unattended, reports land in your inbox, and you're the operator, not the worker.

[00:09:56]But remember, don't start here. Start at level one, get one agent, let it do some real work, let it learn, and then grow [music] from there.

[00:10:04]All right. Now, it's time to land this.

[00:10:06]You now have Air Mason agent installed, configured with your model provider, and you understand the three systems that makes it different. Persistent memory that curates itself, self-evolving skills that learn from your work, and the messaging gateway that puts it in your pocket. Now, if you want to go deeper, you just need to use these resources. [music] And if you want Shawn Humberg's four-level framework, I'll link his full article in the description. It's the best mental model for going from one agent to a full fleet. And Avi Chawla's masterclass, which is the deepest technical [music] breakdown of the memory and skill system I've found. So, drop in the comments what you want to build with this, whether it's SEO pipelines, personal research agents, content distribution bots, automated morning briefings, [music] whatever you want to see. Let us know, and I'll see you in the next video, and until then, keep building and keep experimenting.