The Hermes Agent Setup Everyone's Installing Now transcript

A new AI agent just dropped, and people are saying it kills Open Claw. I've been running Open Claw for over 3 months now. It helped me build a trading bot, get YouTube monetized, and automate most of my business. Today, I'm installing Hermes agent for the first time live on this video. I'm going to show you the full setup step by step, and by the end, I'll tell you honestly whether you should switch. Let's get started.

So, Hermes agent is an open-source AI agent by new researcher Now Research. I'm not sure how to pronounce it, but think of it as Open Claw's competitor. It's the same idea. You run it on a VPS or a local machine, chat with it on Telegram, and it controls your computer and automates your workflow. It's recently been blowing up. This is their website. This is their GitHub as well, blowing up with stars recently. And here's what makes it different.

It has a self-improving loop. Every time it completes a task, roughly every 15 uh tool calls, it pauses, evaluates what happened, and saves it as a reusable skill. It literally gets smarter the more you use it without you doing anything. It also runs on any model you throw its way just like Open Claw, whether it's Claude, GPT, open-source, whatever. You get the spiel. And the install is also a one-liner command, the one I'm showing you right here.

So, without further ado, let's get into it. So, for step one, you need a VPS or a local machine. I'm going to use my local machine, aka uh a Mac Mini that I have here in my office, but any VPS works the same. Step two is for you to go open up your terminal, and literally copy-paste this one-liner like this, and hit enter. There we go. Just like that, it has already started. It will on its own uh install any dependencies it needs.

Like, if it needs Python, Node, Git, you don't have to touch any of that. Okay, and in under a minute, here we are. It's saying, "Let's configure your Hermes agent installation. Ple- Press Ctrl C anytime to exit." Nope, we don't want to exit. We'll walk you through uh model and provider, terminal back-end, agent settings, messaging platforms, and tools. So, let's get started. Let's press enter to begin. Open Claw installation detected.

Would you like to import from Open Claw? Yes, absolutely. I would. Um select provider. And this is, by the way, something I forgot to mention. And something maybe not a lot of people know. If you have Open Claw, it already has an auto-import uh instance where it literally imports your memory, your data, so that you don't have to start from scratch, which is amazing. So, let's let's test it out in a bit. But, with that, select your provider.

I would like to go with Anthropic because, well, we're already paying for the $200 a month subscription, but here you can go with anything. Uh the Claude Pro Max subscription. Select TTS provider. Is it Edge TTS, free, cloud-based, no setup needed? Yes, I'll go with the free one cuz it's free, and I don't think I need anything better for now. Uh reconfigure terminal back-end. Select terminal back-end. Run directly on this machine.

Yes, please. Messaging working directory. Working directory for messaging sessions when using Hermes. This is via Telegram or Discord. This is where the agent starts. Clee mode always starts in the current directory. And what's beautiful here, if you don't know what the hell they want you to do, you can just Open Claw code or any Claude instance, for that matter, and ask it, "Hey, I'm trying to install Hermes agent, and it sent me this.

What do I do?" Ah, okay. So, common choices, your home directory. Since you're running this on your Mac Mini and have things like Open Claw, I'd go with the home directory. And there you go. Enable pseudo support. Stores passwords for apt install. Yes, sure. Pseudo password. Reconfigure agent settings. Yes. Maximum tool calling iterations per conversation. Tool progress display controls how much tool activity is shown.

Uh I want I would like to see all. Context compression automatically summarizes old messages. I'll just keep it at the default. Session reset mode. Inactivity plus daily reset recommended. Reset whichever comes first. Uh inactivity only. Let's start with the recommended, and then as we play around with it, we can always change those settings. Session reset policy. Inactivity timeout. Yeah, sure. Daily reset hour. I'll keep the default.

Reconfigure messaging platform. Nope, it's Telegram, so I'll keep it at Telegram. And there you go. Um browser automation. Ah, these are the tools it's going to install at first. Reload your shell to use Hermes command. So, just say, "Hermes." There we go. I think we're in. Yeah. There we go. Let me expand this a little bit. Hi, Hermes. My God, silly me. So, what I did by telling it to take over the Open Claw instance is it used the same bot token as my Open Claw, which ended up messing up everything.

So, I recommend strongly recommend creating a new bot instance for it. Uh let's call him Hermie. Hermie the Great Bot. All right. Create a new bot. Copy this, and just like that, I'm going to go set that up. What I'm going to do is go back through the exact same steps, but say no here. Would you like to import from Open Claw? No. Uh so, to find your Telegram ID, user ID, message this bot, and it will reply with a numeric ID.

Okay, so the second issue, the one remaining, is Anthropic's uh off subscription doesn't seem to work with Hermes, which is not ideal, if I'm being honest. That's a minus on Hermes' side. So, what I'm going to do is instead install it through Open Router, and give it my API key. Let me go back fetch that. Okay, so I fetched that. I've inputted it. Now it's asking me to choose a model through Open Router. Kimmy K2. Absolutely, Kimmy K2. 21.

Default model set to Kimmy K2. Now, if we run Hermes, it's using Kimmy K2. If I send "Hi," fingers crossed, it's typing a response, so it should be working. There you go. Hi, I'm Hermes, an AI assistant Claude, running through what appears to be a system called Hermes based on my context. You are alive, my friend. Just like that, we got it to set up. Honestly, a lot easier if you don't use your Open Claw instance, and if you don't use your Anthropic subscription.

There you go. It replied, "Haha, yes, I'm indeed alive and kicking. What are you trying to do earlier that was giving you the error?" So, just like that, we have our Hermes agent set up. Now, we can finally start on testing it. Let me show you the thing that makes this different from everything else. I'm going to give it a real task. Every morning at 35 a.m. I don't know, just feeling like that. Find the top three AI stories, top three AI news stories, summarize them, and send me a briefing on Telegram.

That's it. I'm going to type that in. I'm going to copy-paste it, uh and send that in. Just like that. So, let's see what happens. Watch what happens here. It's not just completing the task, it's capturing what it learned and packaging it as a skill in real time. See that? Cron job create times two, terminal, pip install, Cron ter. That's a cool name, Cron ter. Cron job, cron job list. So, it's literally creating a skill for it to be able to improve every single time and not have to start something from scratch.

It does that so that next time I ask it something similar, it already knows how to do it. It doesn't have to start from zero. That's it. That's the self-improvement loop. The more you use it, the sharper it gets. There we go. Just like that. After what? 2 minutes of it running, it is done. It said done. Your daily AI briefing is scheduled. The job ID first run today in about 10 minutes, repeats 30 times. Want me to make it permanent?

Number one, change the time, modify the briefing format. Your first briefing should arrive shortly. That's it. So, after actually installing and running this for a few hours now, here's my honest and real take. Hermes is faster, noticeably. Even though I'm using a slower model for it, Claude v2, and on OpenClaude, I'm using Sonnet as the default model, which is noticeably faster. It should be faster. Hermes still moves quicker.

The self-improving skills are genuinely impressive. OpenClaude doesn't do this natively. You build your skill files yourself. Hermes does it automatically. The more you use it, the sharper it gets without you having to touch anything. But, here's my actual problem with it. Anthropic OAuth doesn't work. I mean, I use Claude for everything. It's the best agentic model out there. And it's what powers my entire OpenClaude setup.

And Hermes, you have to route through OpenRouter to or use an API key to get Claude, which adds a layer of friction and cost that I don't love. That's a real downside for me specifically. And if you're a Claude-first person, like I am, you are going to feel it, too. OpenClaude has 3 months of my life baked into it. My trading bot, my content agents, everything is tuned to how I work. Hermes starts fresh. Now, granted, you can use the migration, the OpenClaude to Hermes migration, smarter than I have, better than I have.

Um so, that is a small caveat. That's more on me than it is on Hermes. So, here's what I've decided to do moving forward. I'll start running both. OpenClaude stays as my main setup for everything that's already working. Hermes runs alongside it. I'm going to let it learn from scratch and see how fast it catches up. Let me know if you'd want to see that as the next video. Now, the question everyone's asking themselves, is it replacing OpenClaude?

I don't think so. Not yet. Not today. But, it is the most interesting thing that's dropped in the space in a long time. And you should be trying it as well. If you enjoyed this video, I have a lot more on my channel. So, make sure to subscribe and hit the bell and do all those things, and I'll see you in the next one.