Claude + Hermes Agent: The #1 Agent Setup That's Replacing Everything (full setup)

本站添加: 2026-05-17

[00:00:00]This is Hermes agent from new research.

[00:00:01]And you're probably thinking new tool, same height, different name, and more lackluster promises. And then 6 weeks later, you have nothing to show for it.

[00:00:10]So, that's why I'm making this video incredibly simple so you start. AI has never been more powerful and it's also never been more scattered. Tons of everything everywhere and at this point it just feels like overwhelm. But in all this noise, there is one setup or workflow that I can confidently say every person that's serious about using AI, which you probably are since you've made it this far, and looking for the lowest hanging fruit right now, that'll give you the most bang for your buck, is a Hermes setup coupled with your model of choice, Claude this in this case.

[00:00:42]So, by the end of the video, my promise to you is a fully functional Hermes agent with two useful skills loaded in, your messaging channel setup, and everyone says Telegram, but I think the extra bit of effort to use Slack is totally worth it. So, you're up and running, you have something tangible and ready to use, which is why I won't be inundating you with systems and workflows you might not even need right now.

[00:01:04]Goal is to show you quick use cases, get you set up, and have you tinkering. Less is more. Over the past 3 months since Open Claude's release to now, and with everything that's happened, I've slowly migrated out of Open Claude and I've replaced it with the Hermes agent. And over the past month of testing, my agent setup has gotten faster, cheaper, stopped breaking, and most importantly, just simpler and a real pleasant time using it. But before we get into it, I wanted to share and give you a few quick things I learned from the mistakes I've made that are genuinely useful that I think wouldn't make this video complete if I didn't at least mention them. With the climate and rate of advancements in AI now, at least as of May 2026, and mind you, I've spent hundreds of hours working in or with AI, AI is changing at a much faster pace than it ever has. Especially much faster compared to the speed it was going at around this time of last year. The point here is do not be rigid and tool reliant. Instead, have the mindset to adapt and change when needed.

[00:02:06]Be like water. Bruce Lee.

[00:02:08]The next one is always have in the back of your head that all of these systems might fail tomorrow and all these providers might just end up pulling the plug. Um and so because of that, you don't need to create and have the mindset of having like an agent agnostic back-end system that is somewhat update proof and kind of like a plug and play so that anything happens, you're just a few prompts or changes away from getting everything back into place and in order.

[00:02:36]And what this might look like, and this can be different for everyone, is having all of your skills, systems, files saved somewhere like in a GitHub repo, second brain, Obsidian, whatever that is. That way your agent can immediately do a check and get all your systems and workflows back in place. I'll be making future videos on these, so make sure you subscribe and tuned in. Lastly, the systems and frameworks you have in place are more important than the tools themselves. Since you'll always fall back to what's easiest, make sure you spend the time to do the hard part early on and prepare your systems. Because if similar to an employee, you train them once and the hope is you reap the returns on the investment in the long run. And the reason I'm mentioning all of this is because Hermes will likely be an integral part of your system structure that you'll be building. So, if you're starting out or maybe already using OpenClaw, I'm going to cut to the chase here and say as of May 2026, skip it altogether or just move on.

[00:03:34]Especially if you've been on the fence.

[00:03:36]You'll be spending more of your time getting it to work rather than working with it. Hermes is an agent harness and in simple terms, it just means it's an infrastructure around an AI model so that you can use it reliably as a personal agent.

[00:03:50]I Hermes because of its persistent memory across sessions, self-improving, its ability to create skills under the hood, and it's just easy to use. It doesn't break and it's reliable. The new research team is clearly doing something right. Plus, I think they have taste.

[00:04:03]Hermes is comprised of five things.

[00:04:05]Starting off with the first one is memory. Memory is basically just the two markdown files, the user.md file, showing who you are, your background, how you talk, what you can and can't stand, who your clients are, things like this. And then you have the memory markdown file, which is the environment, what you're working on, your tools, your setup. Both load into every session by default.

[00:04:25]AI by default has serial memory between sessions, meaning it's fresh every time.

[00:04:30]So, if you don't have these files filled out, you're going to be re-explaining yourself over and over again. And then not only will you be wasting time, you're also wasting tokens and adding additional randomness, especially if you're explaining it differently every time. You're also adding extra bloat and just overall not good. The cool thing is Hermes will be updating these on its own as you talk to it. If you ever want to be explicit, you can always say things like save that to memory or let's pin that, don't do that again. Put that in the user.md file. It'll figure it out and it takes about 2 seconds and can save you hours. Skills. So, anything you do more than twice, it could be a skill.

[00:05:08]It's just markdown files with two parts.

[00:05:10]A small block at the top called the YAML front matter, which is basically just what tells the agent when to use a skill and it's kind of what it's indexing and looking for. And then below that is the actual instructions. The reason why this is so good is because Hermes doesn't pull the whole skill into context, wasting context tokens, until it needs it. You don't pull the conversation with the stuff it's not using or isn't needed. Now, for the part that sold me is if you keep doing something manually over and over again, Hermes will turn it into skill. It'll literally notice and offer to make one for you. If it notices a repeated or important process you're doing over and over again. You can also grab these skills that other people have built, which have primarily been on GitHub, but there's also public skills hub and there's tons of them. But my suggestion though is outside of GitHub and the official website like new research page, I would double and triple check before you download any skills and you can just feed it into an agent to like do a check because they can be malicious. Hermes is already preloaded with I think like 91 skills at the moment and they're continuously adding to it. And during the setup we'll be adding two extra skills, which I think is useful for it to have.

[00:06:22]To download these skills is fairly easy.

[00:06:24]All you have to do is give the URL to your agent and ask to install it and it'll be done in like 5 seconds. The next piece is the soul markdown file.

[00:06:34]Now the soul is the personality. Same markdown file, just no front matter.

[00:06:37]This is what makes one of my Hermes agents blunt and surgical and another one that's a little bit more casual. I have the quality agent, which does a lot of the auditing for scripts, outputs, looking out for AI slop and double-checking work.

[00:06:52]And it's funny because I've assigned that quality reviewer the persona of the character L from the anime Death Note and if you know how locked in L is and how particular, extra cautious and diligent he is, now every time I talk to it or ask it to review something, it'll act in that persona.

[00:07:10]Next one is cron. Cron is just a scheduled task. You can say in plain English, "Every weekday at 7:00 a.m.

[00:07:15]check my YouTube comments or every morning do this, remind me of that.

[00:07:19]Let me know what are the most important things." And when the time hits, it'll run that task and it'll send the results back to whatever channel you wanted it in Slack, Telegram, whatever.

[00:07:30]ChatGPT or Claude, they kind of have it, but not really. They have the routines features, but it's just not there.

[00:07:37]Whereas in Hermes, this will run while you sleep. And lastly, the self-improvement loop. The self-improvement loop ties the other four together. You do the work Hermes watches, you correct it when it's wrong.

[00:07:48]It saves the correction into memory or the user MD. Repeatable stuff becomes a skill. Every conversation gets stored, so when you ask, "What did we decide 3 weeks ago?" it'll actually find it. But this isn't just like automatic magic.

[00:08:02]You have to be active about it. You have to actually correct it. You have to say, "Save this."

[00:08:07]Treat it like training an intern. You spend the time up front and 6 months in, it just keeps getting sharper. So, the return on those early hours is huge. A quick showcase, here's a few of the agents that I made. We have the creative lead. This is the content lead agent that helps me draft and edit scripts and YouTube videos.

[00:08:26]I've even had it help me create a mission control for video editing and this is what it looks like. So, here I can write, I can do edits, I can read through the script, as well as look through the presentation and make edits here as well.

[00:08:41]On the left side here is a pane for all the related files specific to the video that I'm currently working on.

[00:08:47]Basically, all of the stuff in one place that I can easily navigate be custom tailored for how I like to do my video editing. I'll also include the prompt I used to create something like this in the description down below. If you're interested, it won't be perfect and you'll likely need to tweak it just a bit, but should get you more than halfway there. And the other one is just a quality agent.

[00:09:08]Pretty self-explanatory.

[00:09:09]It's a separate Hermes profile that audits my scripts for AI slang language.

[00:09:13]Basically, making sure that it's my voice.

[00:09:15]Now, for this you have to give it context through your own writing, text, or maybe just speak to it. But have it ask you questions so it knows your voice and gets better over time.

[00:09:25]What this agent does is check through the outputs even on the work my other agents are doing.

[00:09:29]Again, basically making sure that it's in my voice and also works as a second reviewer. Looking for mistakes, anything weird, things like this. As we go through the setup process today, since I've already downloaded Hermes, I'm going to instead make a separate agent.

[00:09:44]It's going to be basically the same thing. If it's your first install for Hermes, you're going to see the same setup with creating an agent, similar process. So, before install, there's some additional things that we're going to give your agent during the setup. Two skills and one cron. The superpower skill is the playbook. It's a library of senior engineer best practices that the agent will use for hard tasks. Without it, the agent improvises. With it, it has a system. So, the humanizer skill, that just kills AI slop in any writing.

[00:10:12]So, no more EM dashes and helps drop the output back into your own voice. And the scheduled tasks, aka the cron jobs that we're going to set up, this one is a weekly job that goes back to the Hermes documentation website, pulls everything fresh, the FAQ, the troubleshooting guides, the setup docs, basically things that may not have been covered in Hermes itself from the get-go. So, when you ask your agent how to do something in Hermes, it's reading from the most current docs. And we'll give it a look after the setup in the Hermes dashboard to show you what that looks like when it's all scheduled.

[00:10:46]Here's the chapter map for the install.

[00:10:49]Eight chapters total. If you're on a Mac, you don't have WSL, so skip that chapter entirely and move on to the next one. Quick note, instead of filming my screen live while clicking and typing through different websites, terminals, etc., I took a screenshot of every single step when I was setting up Hermes myself and basically stitched that together into a slide-by-slide walk-through. Every image has a small caption at the top telling you exactly what to do at that moment.

[00:11:17]Like you're seeing here, every slide is going to have a small caption at the top telling you what it's about. And then you'll be commentating over it as I go through it. And there's chapter indicators on the left side, so you always know where you are, and a number on the bottom right corner, so it's easy to pause, rewind, and basically come back to later if you wanted. I find this way cleaner if you're trying to follow along, or if you just want to see a quick overview of the core steps so you can come back when you're ready knowing what to expect and actually tackle this much faster.

[00:11:45]And it's also made in mind for those that want to pause step-by-step and clearly see what's going on. So, feel free to choose your own adventure and go at this at your own pace.

[00:11:54]All right. So, the first slide, again, if you are on Mac, you don't need to do this. Skip to the next one.

[00:12:01]But, if you're in Windows, open up your terminal, PowerShell, and type in WSL install.

[00:12:08]This is what it looks like when it's completed. It's going to ask you to reboot, so reboot your computer after this.

[00:12:13]Once you're back online, you're going to want to install Ubuntu using this command right here, wsl.exe, install Ubuntu. Ubuntu is just another flavor of Linux. It's the most common one that people use.

[00:12:27]After you have that installed, this is what it looks like when it's installing. Then, create your Ubuntu username and password. The bottom here, feel free to put in whatever. I put no.

[00:12:37]And then, once that's completed, go into the Hermes Agent website. On the landing page, you'll see this code to install.

[00:12:44]Copy that. Paste that back in your terminal. Once that's going, it's going to ask you to put in your password that you just made a little while ago. And so, while Hermes is installing, go create a Slack account and download the desktop app.

[00:12:56]Hit quick setup right here.

[00:12:58]Select quick setup. And here's a list of all the inference providers.

[00:13:02]We're going to go with OpenNMT-py for this one. So, second from the top.

[00:13:06]And if it doesn't give you the link, just go to the OpenNMT-py page.

[00:13:10]Set up an account here as well.

[00:13:12]Create a key.

[00:13:14]Take your key and before all this stuff, you probably have to set up your account information, add in your credit card and stuff, but you get here, copy this key.

[00:13:24]Paste it in the bottom. Hit enter anyways after you've copy and pasted it.

[00:13:28]It's there.

[00:13:29]Once you do, this is what it looks like when your API key has been saved. So, now that it's connected to your open router, you can see here a list of all the prices for every million token for the input and output tokens. So, you can see here output tokens generally cost about three to five times more than input tokens.

[00:13:49]We're going to go with Opus 4.7 on this one. So, select that one. Click local here.

[00:13:54]Confirm and continue, and this is where it's going to ask you for your password.

[00:13:57]Put that in there.

[00:13:59]For this one, actually don't mind this title up here. This is just what it looks like after you pasted that command from the Hermes Agent website. There's no password you have to put in here.

[00:14:10]So, here hit local.

[00:14:13]Now, this is where you put in that sudo password that you made a little while ago.

[00:14:18]It's going to ask you to set a messaging.

[00:14:21]Hit Slack.

[00:14:23]And for me at least, it didn't take me to go configure my Slack. It went to the installation complete page. So, we're going to have to go back in there.

[00:14:32]So, to go back in there, open up another terminal. Go into WSL. Type in WSL.

[00:14:39]If you're a Mac, you don't have to do that. Just run Hermes. Uh we're going to start up Hermes for the first time. And so, once that's done, this is what the Hermes Agent CLI looks like. You can actually chat with your Hermes right now from here. But, we're going to set up our messaging channels first.

[00:14:55]So, go back to the terminal. Go back to WSL.

[00:14:59]And type in Hermes gateway setup. Once you do that, select Slack. Once you're on this page, we'll see that it's already written a manifest app into this file path below.

[00:15:11]If you're a Mac, you can click that directly. But, if you are a Windows user, you won't be able to just copy this and paste it into your browser.

[00:15:20]You'll need to add the additional part right here. So, wsl.localhost Copy this whole thing when you're done.

[00:15:27]If you paste it onto the browser, you'll be able to see that manifest app directly. So, you're going to take that, copy it, and then click this link right here, the apis.slack.com.

[00:15:38]You should be taken to this page. If you're new, it should be blank for you.

[00:15:42]For me, this is where all my agents live as apps. Hit create new app.

[00:15:46]Once you do, select from a manifest.

[00:15:50]Choose that workspace.

[00:15:52]Here, make sure you're in the JSON section.

[00:15:55]Select what's already in there by default.

[00:15:58]Paste in your app manifest that you just copied. Hit create.

[00:16:03]After that, go into the basic information.

[00:16:07]At the bottom, go ahead and click generate token and scopes. Once you do that, let's give it a name.

[00:16:13]Add scope. This is very important. And type in connections right, or you can just select it. It should be like the first one.

[00:16:20]Hit generate. Click copy. This is your app token.

[00:16:24]Copy your app token here. Save it somewhere.

[00:16:27]And then, on the left-hand side, click install app. Click install app.

[00:16:32]Install it.

[00:16:33]Hit allow.

[00:16:35]Now, this is your bot token. Copy this.

[00:16:39]Go back to your terminal. This is the token that it's asking for. Paste it in here. Again, it's going to be invisible.

[00:16:45]You won't be able to see it. Hit enter anyways.

[00:16:48]Once you do, it'll ask for your app token from earlier. Paste it in there.

[00:16:53]Hit enter.

[00:16:54]And after that, it's going to ask you for your allowed user ID. This is what's going to make it so that only your Slack account is the account that can message your agents through on the web interface. Go to the top left.

[00:17:09]Click your profile, and you should see some kind of like domain.

[00:17:13]Take that.

[00:17:14]Drop into your browser. Once you do, your user ID should pop up. Mine starts with a T. Take that.

[00:17:21]Paste that back into the terminal.

[00:17:23]And once you do that, it'll ask you to set home. You can do this later. Honestly, just leave it blank. Keep going.

[00:17:30]Here, hit done.

[00:17:33]It's going to ask you for the stuff down here.

[00:17:36]This is just stuff running in the background on restart, stuff like this.

[00:17:40]Honestly, I put yes, but I would put no.

[00:17:42]Hit no. Skip this part. And yeah, this is what done looks like. So, if you want to check, go to your Slack page. Should see that your agent is now in that list.

[00:17:54]And if it isn't in your desktop app or your web interface, what you're going to have to do is send it a DM. So, compose a message.

[00:18:02]For the two section, find your agent.

[00:18:06]And just send an @ to your agent. Say hello.

[00:18:10]Once that's done, we're going to install our skills. So, again, these skills, I'm going to have them in the description down below.

[00:18:17]Basically, just copy that prompt.

[00:18:19]And drop it into the chat. And your agent will do everything. So, should be super fast.

[00:18:25]Next one will be that cron job. Again, the prompt for this one will be in the description below. Paste that in. And like the skills, it should be done. To check on that cron job, we can go into the Hermes dashboard. And to do that, just go into the terminal like this.

[00:18:39]Type in this command, Hermes dashboard.

[00:18:41]It'll spit something out like this.

[00:18:44]You'll take this part here.

[00:18:47]Click on that.

[00:18:48]And once you do that, should take you to your browser. And this should appear.

[00:18:53]This is the Hermes dashboard. It's the command center for all of your Hermes agents. Um you can see every past session, the different models you're using.

[00:19:02]And here's the cron job section. So, we confirm that it is up and scheduled.

[00:19:07]It's this guy right here. So, weekly Hermes refresh.

[00:19:11]And as well as different skills, plugins, profiles, um your API keys as well.

[00:19:20]And yeah, this is a good place to start if something breaks and look around in especially if you're more visual and a little less technical.

[00:19:28]Um here's a Kanban board. So, once you do start a task, you'll see it appear here and kind of go through the workflow until it's completed. Cool little feature here that the new research team included.

[00:19:40]You don't need to live here, but it's a good place to start. This is what my slots looks like. Um I prefer slots because of how it's organized and also because I was already familiar with it from my past jobs.

[00:19:49]And boom, here's tech lead. Oh, and notice here how it's giving you this set home error. What you fix this is just type in space {slash} set home and that should get it done as you see it here.

[00:20:01]Now, I got to go in and tinker with some of the MD files and give them a profile icon. Let me know what persona you think I should give them in the comments below. Everything's going to be in the description and if some of this stuff was kind of confusing for you, the terminology, check out my 19 concepts video where I run through some of these core concepts. It's kind of like a quick crash course on some concepts that you'll likely see in the future. If you learned anything in this video and was helpful at all, feel free to drop a like and if you want to see some more stuff like this, leave a comment or subscribe.

[00:20:32]See you.