§ 01 · The Hook
The bait, then the rug-pull.
Nate Herk opens with his live Hermes instance already running — voice replies in Telegram, cron jobs firing on schedule, a YouTube comment bot that knows his entire channel history. The promise: you can have this too, no Mac mini required, on a $6/mo VPS, in under an hour.
What the video promised.
stated at 00:15 "By the end, you're gonna understand exactly how to actually get the most out of this super powerful AI agent." delivered at 58:00
Where the time goes.
00:00 – 03:19
01 · Live demo intro
Shows live Hermes in Telegram: cron jobs, voice replies, hyperframes video the agent self-generated. Hook and promise established.
03:19 – 07:36
02 · What is Hermes + comparison to Claude Code and OpenClaw
Open-source MIT from Nous Research, 140K GitHub stars. Mental model: Claude Code = desk work; Hermes = on the go. Comparison breakdown of all three tools.
07:36 – 16:20
03 · The Five Pillars
Memory (user.md + memory.md), Skills (YAML front matter + Skills Hub 520+ community skills), Soul (soul.md personality), Crons (natural language scheduling, fresh isolated sessions), Self-Improving Loop. Each shown as Excalidraw diagram.
16:20 – 25:30
04 · VPS setup on Hostinger
KVM 2 plan, Ubuntu 24.04 LTS, Docker one-click install vs root install. Nate strongly recommends Docker for agent isolation. Shows Claude Code VPS-agents management project pattern.
25:30 – 33:00
05 · Onboarding + Telegram connection
Choose OpenAI Codex (cheapest, uses existing ChatGPT subscription). BotFather bot creation. Set allowed user ID. Launch CLI. First voice message auto-builds user.md.
33:00 – 43:13
06 · GitHub backup + first cron job
Secure API key flow: hermes config set in VPS terminal, NOT chat window. Creates private repo. Auto-generated nightly-github-sync skill + midnight cron from natural language request.
43:13 – 46:20
07 · CLI vs Telegram mental model
CLI = cockpit (deep work, slash commands, context visibility). Telegram = remote control (quick tasks, voice, scheduled reports). Same agent, same brain — Telegram hides context management.
46:20 – 50:30
08 · Best practices + security
Treat agent like an intern: narrow API scopes, own accounts per agent, VPS firewall via Hermes research, nightly security audit cron. Context compression demo at 167K tokens.
50:30 – 58:22
09 · Scaling multiple agents
Each agent in own Docker container. Decision tree for when to spin up a new agent. Bad pattern: one mega-agent with all keys. Good pattern: Personal Manager Hermes routing to specialist agents.
Visual structure at a glance.
Named ideas worth stealing.
07:36
list
The Five Pillars of Hermes
- Memory (user.md + memory.md)
- Skills (skill.md with YAML front matter)
- Soul (soul.md personality)
- Crons (natural language scheduling)
- Self-Improving Loop
The five architectural concepts that make Hermes more than a chatbot — together they create a persistent, improving, proactive AI teammate.
Steal for Any agent setup talk, CLAUDE.md design, JoeFlow agent philosophy content
08:11
concept
Memento Mental Model for AI Agents
AI agents wake stateless every session, like the movie Memento. The context files (user.md, memory.md, CLAUDE.md) are the sticky notes on the wall that let it function.
Steal for Explaining CLAUDE.md to newcomers, any onboarding content about context management
43:50
concept
CLI = Cockpit, Telegram = Remote Control
CLI gives full visibility: slash commands, context window, session management. Telegram is the mobile interface — same agent, less visibility. Don't vibe-code complex apps from Telegram.
Steal for JoeFlow positioning: dictate on the go, do deep work at desk
50:30
model
Agent Scaling Decision Tree
- Needs different secrets/tools? -> new agent
- Needs separate long-term memory? -> new agent
- Is it ongoing repeated work? -> new agent
- One-off task? -> keep in personal main
Simple rule for when to spin up a new Hermes vs. adding to your existing one. Start with one, scale naturally when you hit criteria.
Steal for Multi-agent architecture talks, agent fleet planning content
46:40
concept
Least Privilege for Agents
Give each agent only the credentials and tools needed for its specific job. Marketing agent doesn't need QuickBooks access. Finance agent doesn't need social posting.
Steal for Security-focused content, responsible AI agent deployment
Lines you could clip.
00:03
"Hermes agent is one of the most powerful AI agents that I've ever played with."
04:22
"Cloud Code is still my daily driver. That's where I do 90% of my knowledge work throughout the day."
08:11
"If you guys have ever seen the movie Memento — that's kind of like how agents work."
43:50
"CLI is the cockpit. Telegram is the remote control."
58:10
"This isn't a tool you finish setting up. It's a teammate that you keep using and you keep training."
How they spent the runtime.
Hook length199s
Info densityhigh
Filler8%
Sponsor blocks
- 16:30 – 17:30 · Hostinger
Things they pointed at.
03:20toolOpenClaw
How they asked for the click.
57:30
link
"The link for the free resource guide is down in the description. You'll go into the free school community, click on classroom, click on all YouTube resources."
CTA delivered twice: once mid-video before setup section, once at end. Free resource guide positioned as a companion doc you can hand directly to your Hermes to execute the setup. Smart integration.
Word for word.
HOOK opening / re-engagementCTA the pitch metaphor analogy
00:00HOOKHermes agent is one of the most powerful AI agents that I've ever played with. So in today's video, I'm gonna take you from absolutely nothing to being able to get one set up. And by the end, you're gonna understand exactly how to actually get the most out of the super powerful AI agent.
00:11HOOKThis is the landing page for Hermes agent. It is an agent that grows with you, so it has, like, sort of that self improving loop with skills and stuff. So it's very, very cool.
00:19HOOKAnd it's a lot easier to set up than you'd think. We're not gonna have to use a Mac mini or anything. I'm gonna show you how you can just set one up on a private server.
00:24HOOKNow this thing out of the box is already super powerful. If I go to docs and then I click on skills, you can see that there's, like, 684 total skills, but there's 91 that are already basically just built in once you install Hermes.
00:36HOOKSo for example, you can see that it generated me some Excalidraw diagrams, which we'll look at in a sec, but I didn't have it install any sort of Excalidraw skill. It just had that already. And I never gave it a transcription skill or a voice skill, but take a look at this.
00:48HOOKHey there, sir. Can you just go ahead and tell YouTube who you are, what you do, and what some of your crons are? By the way, when I say cron, that basically just means, like, an automation.
00:56HOOKIt's a scheduled automation. And don't worry, we're gonna dive into all of that later in the video. I'll also talk about Hermes compared to Cloud Code, compared to Open Claw, the differences and why in this specific video I'm going with Hermes.
01:07HOOKSo you can see that it shot back an actual voice note as well as text. So this is a full minute and sixteen seconds. I'm not gonna play the whole thing, but let's take a quick listen.
01:17HOOKI'm Nate's AI assistant running on his own infrastructure. I'm not just a chatbot in the browser. I can use tools, remember preferences, write reusable skills, run scheduled automations, search past conversations, work through Telegram, and help manage real workflows.
01:26HOOKThe way I Alright. Thank you so much, Hermes. But, anyways, take a look at some of these crons that my Hermes is running.
01:31HOOKA daily AI news briefing, which is posted inside of my school community, YouTube comment monitoring. So if you guys have noticed on my YouTube videos lately, I've had an AI agent that has access to the transcript and knowledge about me, and it's been responding to your guys' comments. And that is this Hermes agent.
01:44HOOKSchool community engagement, morning business summaries, server checks, research reports, follow-up reminders, and that's just some of the crons that I have this agent working on. And if you guys have seen my videos on my channel about hyperframes with Claude code, basically to edit videos, I wanted to see if Hermes could do that.
01:57HOOKSo I said, hey. Can you make me a video using hyperframes about what Hermes agent is, how you work, how you remember things? It ran all of these different things.
02:04HOOKSo it has a skill called creative. It had a skill called manim video. It was searching for hyper.
02:09HOOKIt ran all these terminal commands, and it also you can see here, it's using vision to analyze the actual video to see how it turned out. Now its first pass wasn't great. It didn't even use hyperframes.
02:19HOOKI'm not exactly sure what this used. But as you can see, some of the spacing is off. It's not amazing,
02:23HOOKbut, you know, it's not terrible for the fact that I just said, hey. Make me a video. So basically then I said, What tool did you use?
02:28HOOKI wanted you to use hyperframes, so it had to look into it. It did the research on its own. It asked me if it could install hyperframes.
02:34HOOKI said yes, and then it comes back with a video that's actually much, much better. So I'm not gonna play the whole thing, but here's the video that it actually came up with. It looks a lot better.
02:41HOOKThe spacing is a lot better. It just has these diagrams that actually, like, don't overlap, and they're not going out of bounds. So think about this.
02:48HOOKOne natural language request. It did the research. It wrote this, and all I said, as you guys can see, is, hey.
02:54HOOKMake me a video about what Hermes agent is and how your memory and skills work. It should feel fast paced and exciting.
03:00HOOKSo that's one mindset shift here is if you are confused about anything to do with Hermes, Hermes probably understands it the best, and it can also look up its own documentation. So just ask it, hey. Can you do this?
03:10HOOKIf you see something cool on x, grab that x post, give it the link, say, hey. Read this, and then help me implement it. It's really going to be your best friend here by just brainstorming,
03:19HOOKand then you tell it to go figure out how to do it. Okay. So that was just a quick demo about what Hermes agent looks like when I'm using it through Telegram.
03:25HOOKYou can use it through tons of other platforms as well. Let's just actually dive into this video here. Hermes agent from zero to your own assistant.
03:31HOOKOkay. So what is Hermes agent? It is an open source AI agent from Noose Research, and it is an MIT licensed, like I said, open source project.
03:39HOOKRight now, has a 140,000 GitHub stars, and that is growing really fast. It's one the fastest growing open source projects on GitHub. It runs on your own infrastructure, whether that is a Mac mini, a laptop, a VPS.
03:49HOOKIt can run inside a Docker container, wherever you wanna put it. It can even run on Android via Termux. There are tons of different messaging platforms.
03:56HOOKI'm gonna be showing you guys Telegram today. You could also do Discord, Slack, WhatsApp. You could even do iMessage if you wanted to connect it to that.
04:02HOOKAnd really the big thing that got me interested in trying it out was the self improvement over time by writing its own skills and updating those. And it's kind of built on top of five main pillars, which I'm going to talk to you guys about in just a sec here. But before we get into that, I wanted to cover Hermes versus Cloud Code versus Open Claw and kind of even like Codex too.
04:18HOOKSo this is just my comparison of the way that I compartmentalize them in my head. So Cloud Code is still my daily driver.
04:25HOOKThat's where I do 90% of my knowledge work throughout the day. But there's a clear distinction in my mind between the way that I'm gonna use Cloud Code and OpenClaw or Hermes. So this is obviously Anthropic's coding assistant.
04:35HOOKIt lives in your terminal next to your code, and you basically sit there and you drive it. You could enact like dispatch or remote control to use it on the go, but honestly, I don't really do that too much. The way I think about Cloud Code is when I'm sitting down at my desk or I'm on my laptop and I'm doing work.
04:50HOOKNow OpenClaw is where I started to play around with it. Did the training video if you guys saw that with OpenClaw. And the way that I started to think about this was I'm not gonna use OpenClaw or Hermes to sit down and do, like, my knowledge work and my coding.
05:01HOOKI'm gonna use OpenClaw and Hermes when I'm on the go, when I wanna be on my phone and be able to set up crons really quick and have everything kind of just be managed right there in Telegram where I can talk to something and it wakes up immediately and responds to me back. And it's truly been a game changer for being able to go on a walk and still do work or, you know, be out and about.
05:18HOOKThis was created by Peter Steinberger. He then joined OpenAI, and OpenClaw is still an independent, once again, also an open source project that has over 350,000
05:27HOOKGitHub stars now. There's a much larger team around OpenClaw compared to Hermes, and they are also doing frequent updates.
05:33HOOKAlso, NVIDIA built NemoClaw on top of OpenClaw as a separate enterprise stack. Now Hermes and OpenClaw may seem kind of similar when you just kind of take a first glance, but there are a lot of differences. Hermes is also lighter, faster, focused on self improvement, and they've come out and say like, hey.
05:47HOOKYou know, this is built for people that wanna tinker with open source models, Quen, Llama. I'm not necessarily using Hermes right now with open source models, but definitely something that I'm gonna start experimenting with. But one of the main reasons that I started kind of switching over to Hermes was my OpenClaw was just kind of breaking a lot.
06:01HOOKThey would push a lot of updates and changes, and sometimes it would just like crash my OpenCLAW, and I'd have to get in there and fix some stuff. And Hermes doesn't seem to do that as much. Fingers crossed.
06:09HOOKBut a lot of people are using these altogether, and I'm definitely using Cloud Code with Hermes together for sure. Because, I mean, if you think about what are your coding agents actually working in, they're working in some sort of directory, which is just a file structure, a folder structure. And all of that, we wanna sync to GitHub.
06:25HOOKSo if you have a GitHub repo of all of your knowledge, all of the business context, all of your skills, you can pick out any of these agents, even codecs, and just plop it on top of your GitHub repo, and now you can play with all these different tools and see how they interact. There's just a little bit of difference sometimes with terminology, whether that's like a claud.md or an agents.md or, you know, a couple little tiny things, but each agent understands its own terminology.
06:47HOOKSo if you say, hey. Take this repo and make sure you can use it, it should be able to make all the changes for you very quick. And I'm gonna show you guys a way that I use Cloud Code to help me manage all of my Hermes agents and OpenClaw agents, which makes me stay way more organized.
06:59HOOKI never forget things. And trust me, it's definitely a game changer. So anyways, before we jump in and we start doing the install and the onboarding to Hermes, I just want you guys to understand some of these, you know, main concepts to think about, which are the five pillars.
07:12HOOKAnd what I did is I actually copied this exact five pillar structure, which by the way, my Hermes agent helped me think of. I copied this, I pasted it into Hermes, I scroll up a little bit here, and I said, hey, can you just use the Excalidraw skill to generate diagrams for all of these and make sure that all of this information is correct?
07:29HOOKIt gave me a ZIP file right here, and I took that ZIP file, I put it into Excalidraw, and let's now take a look at these five pillar diagrams. Okay. So the first pillar is memory.
07:39Memory is the small, durable context that Hermes should carry across sessions. So there's two main files to be thinking about when it comes to memory. The first one is the user dot m d.
07:48Who you are, your style, your preferences, and things that you don't like. The second one is the memory dot m d. This is like the environment, the projects you're working on, some of your business context potentially, and these two files get loaded at the session start so that it always kind of knows what's going on.
08:03Because the way that you wanna think about AI in general is that it wakes up stateless, meaning it wakes up with basically no memory. If you guys have ever seen the movie Memento, that's kind of like how agents work.
08:13So it's your job to make sure that the context that gets loaded in, cloud dot m d, user dot m d, memory dot m d, agents.md. It's your job to make sure that those files are pretty holistic so that every time you wake up an agent, you don't feel like you're repeating yourself. And don't worry.
08:26These files, Hermes agent understands, and it's automatically gonna start extracting things about you and extracting things from your projects and putting these files together so you don't actually have to, like, manually consciously think about it. As you can see, the session would start. You would go ahead and start talking, building skills, doing knowledge work.
08:42And as you're starting to add tools and as you're starting to give more info and make changes, it's going to automatically come back and update these files for you. Now that doesn't mean to just be completely oblivious to it. You still wanna say, hey, by the way, chuck that in the memory.
08:53Or hey, make sure you don't ever do this again. Throw that in the user.md. Stuff like that.
08:57And you can see the Ascala jaw diagram wasn't perfect. I had to expand that a little bit. But here's some beginner nuance.
09:01Save durable preferences and facts to memory. Use session search for old conversations. So it's basically able to store all of your sessions into a SQLite database, and it can go search through those.
09:11And do not store secrets or temporary task status. So we'll talk about API keys and the way that you should be putting them into your Hermes agent responsibly once we get into the setup. So that is the first pillar, memory.
09:21The second pillar we have here is skills. So skills are procedural memory, reusable playbooks for how to do a task well. If you guys have already been working with Codex or ClaudeCode or all these other things, you probably understand skills, but I'll give you the real quick lowdown.
09:34Basically, of a skill as a recipe. Someone asks you, hey, can you make me some chocolate chip pancakes? You wanna pull up a recipe, and then you're gonna follow that recipe to a tee.
09:41That's And how your pancakes turn out the same and, you know, yummy every time. Otherwise, if you were just going off of memory of how to make the chocolate chip pancakes, sometimes they might be a little more burnt than the other times, sometimes they would have less chocolate chips than other times. You just want them to be consistently done in the same way, and that is your skill or your recipe.
09:59So all of these skills are in a file called skill dot m d. They have a YAML front matter, which is basically just like a little front matter that that tells the agent, hey, this skill does this, so use it for x, y, and z. And that's basically a concept called progressive disclosure, which helps make sure that you're not loading full skills,
10:15full context bloat into a session if you don't actually need to use that skill. So Hermes will understand the use case for skill. It will then invoke the skill and read it, and then it will ship that information into the session and then
10:26invoke the skill and do what you need. And what's cool about the Hermes agent is that if you're doing things frequently, if you forget, hey, let's build a skill out of this, it will analyze conversations, it will analyze your workflow, and it will turn things into skills.
10:38And then of course, as you use skills more and more, if you're giving feedback, it's going to update those skills as well. As you can see, these Excalidraw diagrams keep messing up right here, which is basically just the beginner nuance. So memory equals what to remember.
10:50Skill equals how to do it again. Hermes can create or patch skills after real work. So super cool.
10:56And, of course, there's a skills hub. So any skills you've already built can be used in Hermes, or you could go to the skills hub and you can see that there's over 520 community skills.
11:05There's different categories, so you can definitely search through here to see what you can add to your Hermes to make it even more powerful. It looks like there's actually 16 right here that are anthropic official skills.
11:16We have canvas design, front end design. We have a skill creator skill. So you can pull all of these into your Hermes agent super, super easily.
11:24You just have to basically run this command, or, you know, you could drop in this URL to your Hermes agent and say, hey. Install this skill right now.
11:31Alright. So that's pillar two. Pillar three is the soul.
11:34Now soul m d basically shapes the assistant. This shapes your Hermes agent. So that if you have six different Hermes agents, they all have basically a different
11:42vibe. Some can be concise. Some can be rude.
11:45Some can be I don't know. But anyways, the soul.
11:48Md is another markdown file that gets put into the context of the agent, and now it's able to just have a bit of a personality. So if you're also letting other people interact with your Hermes, they will feel the personality. If your Hermes is, you know, commenting on YouTube videos like mine, there will be some sort of personality.
12:04The one in my YouTube comments, I told it to be, like, very sarcastic, but not rude. And by the way, guys, it's not super super important that you need to, like, know what a skill file looks like or know exactly what a markdown file looks like, but don't get intimidated. Here is a skill.
12:17So this is skill dot md. This one's called generate image. So everything in between these two lines up here, that is the the YAML front matter that I was talking about.
12:25This is what your agent reads in order to understand, okay, should I use this skill or not? If it then decides, yes, I should use this skill, then it's going to read all of this other stuff, which is just markdown. And markdown just basically means that it's using, like, headers and bullets, and it's just a way for agents to read structure
12:41within a block of text. Like, this means bold. Anyways, that is what a markdown file is, that's what a skill file looks like.
12:47So a sole file would look like that, but there wouldn't be that YAML front matter. But as you might have guessed, the sol file will also evolve over time based on the feedback that you're giving your Hermes. So what is pillar number four?
12:59This is where it gets super cool, and this is one of the main value props for me over something like Cloud Code is the fact that I can just say, hey. Spin up a cron job to do this at this time, and it just does it. Cloud Code, obviously, you have your routines.
13:11You have your loops, but that usually requires you to leave some sort of infrastructure on. Unless you're using, like, the that new Cloud routine, but you're only limited to 15 of those, at least on the the max plan, 15 of those a day. So once again, Hermes doesn't replace Cloud Code for me, but it's kind of my on the go spin up things really quick, and that's why I love it.
13:29So cron's turn Hermes from reactive into a proactive scheduled automation, and you're still getting that full agentic loop if you want it. So you could say in natural language, hey. Every morning at 6AM,
13:39I want you to do x, y, and z. It will go ahead and use a skill and use its tools to create that cron job. And then when that time hits, it will basically invoke, a fresh isolated session.
13:48It doesn't inherit any of the context that you're currently, you know, having that conversation about, and then it will just run that skill. After that happens, it will send that result back to the original chat, and it will maybe update any local files or do whatever it needs to do based on the skill requirements. So some useful pieces of advice down here.
14:05Context underscore from is to pass one job output into another. Workdir is, you know, work directory, and it runs tools from a project folder. And then you can do this flag for no agent, which is just a script.
14:16So, hey. I just want you to run this Python script. I don't want the agentic harness loop inside of that.
14:21I just want the script to be ran. So if you think back to the WAT framework, workflow, agent tools, you'd basically just be deploying the workflow on something like modal. You're not deploying the agent as well.
14:30So safety nuance, cron sessions cannot recursively create more cron jobs, so the prompts need to be self contained. And if any of this is starting to feel a little bit overwhelming, I just want you to understand this terminology so that when we hop into setup and the onboarding, it all clicks a little bit better.
14:45So just stick with me. Alright. And then the last pillar here is the self improving loop.
14:49Hermes improves when useful experience gets persisted as memory, skills, and searchable history. So if you think about the loop like this, you do the work, the agent learns. You save things to memory or to agents dot m d or to, you know, user dot m d.
15:02And then you turn those repeatable steps into skills or your preferences into more memory. And then the agent's able to search past sessions when old context matters, and then you basically just go in that loop over and over. So the more you use your Hermes agent, the better it's going to get and the more it's going to understand you.
15:18So the nuance here is that automatic does not mean magic. The loop works best when the user corrects Hermes, asks it to save things to memory, and lets it create and update skills after you've done some complex work. And there is one more kind of honorable mention,
15:31which is the context file. So agents dot m d, if you're using codex, you know what that is. If you're using Cloud Code, this is the Cloud dot m d, which is kind of just like the overall project goal, kind of like the structure of the project.
15:42So this is something that you're more so gonna use if you're, like, coding in different projects with Hermes, and I would say more so if you're in, like, the terminal using Hermes. And in today's video, I'm not gonna focus too hard on the terminal. I am gonna talk about the difference between using it in the terminal or using it through Telegram or whatever other channel you use.
15:59CTABut this is not gonna be a deep dive on Hermes in the terminal because once again, any terminal style work that I'd be doing, I would just be doing that in Cloud Code. That's the way that my workflow currently exists.
16:09CTABut anyways, this honestly works pretty similar to the way, like, the memory or the soul file works, but those are all global. And this one's more of like a local project. This is what we're working on in this contained environment.
16:20CTASo anyways, hopefully, you guys aren't too bored yet. Let's actually go ahead and get your guys' hands on. I am gonna say before we jump into this, all of this video, I'm gonna have broken down into a resource guide, which might even be helpful for you to just give your Hermes agent the document and say, hey.
16:35CTAHelp me get all this set up. But if you wanna access that free resource guide that breaks down everything that we're gonna talk about today, that will be in my free school community. The link for that is down in the description.
16:42CTAYou'll go into here. You'll click on classroom. You'll click on all YouTube resources, and you'll be able to find every doc, skill, GitHub repo, everything I've ever dropped for free on YouTube right in there.
16:52Okay. So now let's get into the setup of Hermes agent.
16:55Alright. So the way that we're gonna be doing this is on a VPS, which stands for a virtual private server. Now I'm gonna be using Hostinger for my Hermes agent.
17:03I have been using Hostinger for hosting n n n and for OpenClaw and for Cloud Code. So this is my VPS provider of choice. There's a link in the description if you guys wanna go here.
17:11You can see there's also basically, like, a one click install for Hermes agent when you spin up one of these VPS. So very, very cool. Now the first thing you have to do is choose the plan.
17:20So on here, can see KVM one, two, four, or eight. I'm just gonna go ahead and start with two for now. This basically just changes, like, how much CPU and RAM you have and your bandwidth in your server.
17:30So you could definitely start on one, and if you need to just scale up, you can scale up later. But I'm just gonna go ahead and pick KVM two, and then I'm just gonna go ahead and click on deploy. So you have to choose your period.
17:41So twenty four months, twelve months, or one month. I think that you should just probably go for the annual at least because you're gonna save more money. But also, you know, you pay about a $100 and then you have this just set up forever.
17:51And you can also or for a year, I guess. But you can also deploy multiple different Hermes agents or even multiple OpenClaws and Cloud codes on your VPS as long as you can support the RAM and the CPU. And if you choose an annual plan, so twelve months or twenty four, you can use code NATE HERC, and you can save an additional 10%
18:09on that plan. So you can see right here, it says Hermes agent auto deploys with your VPS. You can also get daily auto backups, and then you're gonna choose your server location.
18:17And then once you've made your payment, you just have to go ahead and get started with setting up your VPS. So you'll choose your server location. Click on next.
18:24This is where you can see the actual OS that you wanna use. So I'm gonna do Ubuntu, and I'm gonna do 24 o four LTS.
18:29What And what else you can see here is that there's tons of different apps that you could also deploy, NNN, NemoClaw, and this is where you could come in here and you could search Hermes Agent as well. And if you do wanna do the one click install, then go ahead and do Hermes Agent.
18:40But if you wanna do it on the root of your VPS, then just stick with me here. I'm gonna do a breakdown of the differences there in just like a minute. So if you wanna wait until you see that, just wait for a sec.
18:49So you're gonna have to go ahead and create a root password. I'm gonna go ahead and click next. If you forget that later, you can obviously just
18:55regenerate it. So it's not a huge deal, but obviously, you probably wanna remember that. You can add a malware scanner for free, so I'll just check that box and hit finish setup.
19:02And now this will take just a couple minutes to actually spin up your VPS. So while we're waiting for that, let me talk about the difference between setting it up kind of with the one click or setting it up directly at the root of your VPS. Okay.
19:14So a VPS is basically just a computer in the cloud that you will be renting from Hostinger. You will get an IP address and you will get a password in order to basically SSH in, which means you were just, like, getting into that virtual private computer so that you can manage the files and install things, stuff like that.
19:30So if you install Hermes directly on the VPS, it will be at the root level. And if you install it using the one click Docker image, it will be in a containerized Docker environment within your VPS.
19:40So I know that might make, like, no sense at all, but here's a quick visual. Your VPS has data. It has, like, services.
19:46It has other files, and then you could also put the Hermes agent right there, or you could have all of your files and stuff, you know, locked into the root of your VPS, and then you can spin up an individual container for all of your different Hermes agents or OpenClaw agents within the actual different Docker containers.
20:02So for the sake of the video, I'm gonna be doing the Docker container approach because it's just a one click install and it's much simpler, but even the root VPS is very, very simple. Now here is something that I am very, very strong on that some people might disagree with me and think that it's over engineering, but I think it's like, why would you not do this?
20:19For all of my VPS agents, I have created a my own Cloud Code project to help me manage them. So right here, can see this is called Upit agents. If I click on VPS agents, you can see that I have my bull, which is my trading bot.
20:30I have my main Hermes. I have my Upit OS, and I have Klaus, which was, my main personal assistant. And so for each of these, I can see my passwords and my environment variables, and I can see, you know, like, what's the IP address of this VPS, and how do we have this set up in the docker or at the root, and it has an information about security and integrations.
20:46And basically now, I have one clean place to manage all of my different agents I've got a lot of different VPSs running, and I don't wanna forget my passwords, and I don't wanna forget, like, which agent is on which server. So I just have this project set up. So I would definitely recommend you guys do this.
20:59And I'm actually gonna do this live with you guys here to show you what I mean. So alright. We are gonna go ahead and set up a new VPS, and we're gonna do a new Hermes agent.
21:07So in the VPS underscore agents folder, create a new subfolder called YouTube Hermes. And then we're gonna go ahead and set up, like, you know, like, the passwords and the configuration stuff just so that you can help me make sure we maintain this project well. Now the reason I like to do this is because do you think that I like
21:23understanding VPSs and terminal commands and CLIs? Not at all. And I'm not very good at it.
21:28So who's better at that than me? Hermes agent and Cloud Code. And sometimes if you're running random commands and you don't know what you're doing, your Hermes agent might, like, shut down.
21:37And now I have Claude code to help me boot it back up if I need to. And this may sound scary and overwhelming, but trust me, it is so, so simple. This is just a good best practice to keep yourself organized.
21:47Okay. So switching back over to our Hostinger dashboard, you can now see that we have our VPS ready. So if I click on manage VPS, this opens up our main dashboard.
21:56Couple things to look at. So first of all, have our root access. This is what we can actually give to Cloud Code.
22:03So if it needs to root into our VPS and, like, look at our passwords or help us fix things, it can do so. It would just need the password as well.
22:12So here's where you can change that root password. Down here, you can see your current plan. You can see when it expires.
22:17You can see your, you know, your analytics, and server data will pop up, and it will show you if you need to, like, upgrade to a higher plan. And also you can see on this left hand side, these are all the different servers that I've got running, which is why I want a Cloud Code project to help me keep that organized. So the other thing you can do is you can change the host name.
22:31If you come into here, you just have to have it end in a dot something. So if I just do YouTube Hermes
22:38and then I do dot VPS, I should be able to get that to go through. And now this host name in my dashboard will change to that so I can just keep myself a little bit more organized.
22:47Okay. So the two methods. Right?
22:49If you wanna do this at the root, you would basically just open up a terminal or Hostinger gives you a terminal right here, which means I click on this button. This opens up a terminal that's already SSH'd into our project. You can see it's root at YouTube dash Hermes, and this is where you would just go ahead and go to Hermes,
23:05and you would just run the install command. So you would come down here and you would install with this one line command, and then you'd go ahead and do Hermes setup and start configuring your Hermes agent. But like I told you guys today, what we're gonna do is the one click install.
23:16So we would go over to our Docker manager, and this is where we'd go ahead and click install. And remember how we have our main server, but if we wanted to spin up a bunch of different Docker containers inside of this server,
23:26we could do so. And that's how you could keep, like, different Hermes agents in here and kind of keep them separate. Okay.
23:30Well, this says it's gonna take ten minutes. I doubt it. Okay.
23:32Yeah. It just finished up. So I'm gonna click compose.
23:36I'm gonna go to one click deploy. And then when this loads up, this is where I will search for Hermes, and I will click on select. Now here is where you have to set an admin username and an admin password for your Hermes agent.
23:47I'm just gonna leave this as default. I'm gonna copy this password, and what I'm gonna do is I'm gonna go to my ClogCode project, and I'm gonna save the admin password and username into this new one. So right here, you can see it made this one called YouTube dash Hermes.
24:01And what I wanna do is I'm gonna add a new file in here, call this one the dot ENV. And then in this dot ENV, I'm basically just gonna do admin underscore username
24:11equals and then admin underscore password
24:16equals. And then I'm gonna paste those two things in here. So this is where I can keep myself organized with that.
24:21Alright. So I've saved that to my dot ENV file, and now I can click deploy. This And is gonna spin up that container.
24:27And now if we wanna access this, all we're gonna have to do is click on this little button right here that it will give us, which will basically, like, put us into that container. And we can chat with Hermes, we can do the onboarding, we can do everything in there.
24:38But this main terminal button, that is gonna take us to the root of our VPS, not inside of this Docker container. I hope I'm not losing you guys. I know that it may seem like I'm an expert at this stuff, but truly, I'm all self taught.
24:49I ask Hermes, hey. Explain this to me. I ask Claude Cote, hey.
24:52Explain this to me. And that is how I've learned all this stuff. So it's really not too bad.
24:56Okay. So while that's getting set up, let me go back into our project that we manage our VPS agents on. Let's answer some questions.
25:02So has the VPS already been provisioned on Hostinger? I'm gonna say, yes. It's already been provisioned.
25:09What's the primary scope for this YouTube Hermes? I'm just gonna say other for now. You don't really need to know that yet.
25:14Right now, we're just setting this up as a demo. Alright. Telegram bot.
25:17We will do a new bot via Botfather. Yep. And for the LLM provider, we will be using Codex GPT OAuth.
25:24So I will show you guys all of that as we get onboarded. By the way, the speech to text tool that I'm using right here is called Glido.
25:30It is our speech to text startup. I fully transitioned over from Whisper to Glido, and I am now official member of the Glido team. It's faster.
25:37It's completely private, and it's way more gentic. So check it out. Link in the description.
25:41Okay. So now you can see that this Hermes agent thing is set up. All I have to do now is click on open, and this is where you have to go get the admin username and password that you just saved.
25:50So go grab that and put it in. And once you put that in, you basically just start the onboarding right away. So we're gonna do our quick setup, so I'm just gonna go ahead and hit enter.
25:57We then have to choose an inference provider. So as you can see, there's tons of different options, like a ton of different options. But what I wanna do is I wanna use OpenAI Codex.
26:04This actually lets you take your ChatGPT subscription, so $20, a $100, $200 a month, and use that inside of Hermes agent instead of API keys. So it's gonna be the cheapest option by far besides using, like, you know, an open source model. So I'm going to choose OpenAI Codex.
26:20Now what happens is it makes you go to this URL. So I'm gonna click on that. It has you sign in to your ChatGPT account, and then you have to give it access.
26:27And then you have to go back into the terminal. You're gonna grab this, what is that, a nine digit code and copy that, and then paste that into there.
26:34Hit continue. And when you go back to your VPS, you should be fully signed in, and it should authenticate. There you go.
26:40Log in successful. We get to choose a model now, so I'm obviously gonna choose GPT 5.5. Now we're gonna set up our channel.
26:46So set up messaging now. I'm gonna choose that. We're gonna go ahead and hit space to hit Telegram.
26:52This is where you could also choose some other things if you want, but for now, I'm just gonna go Telegram and hit enter. And now it asks us for a telegram bot token. So what we have to do in our telegram is we have to start a new conversation with the bot father to create a new bot.
27:05So here I am in the bot father. I'm going to do slash new bot. It asks us for a name.
27:10This is gonna be YouTube Hermes, and then it asks us for a bot username. So I'm gonna try YouTube Hermes.
27:16Oops. YouTube Hermes bot. Does that work?
27:18Username is taken. Okay. An absolutely
27:20super ugly name, but that works. And now this is the token right here that you're going to need to copy and you're going to give to your VPS down here.
27:30So I'm gonna go ahead and paste that in. It might not actually appear. Sometimes it just does that in the terminal, but I'm gonna go ahead and hit enter anyways.
27:35So now that token has been saved. And the next thing you have to do is allow a certain user ID. So right now, I only want my Telegram account to be able to talk to our Hermes.
27:44So we have to go get our own user ID. It gives you the steps right here. You have to message the user info bot, and it will give you your number.
27:51So once again, if you go back into Telegram and just search for user info bot, it will look like this. This is where it gives you your ID. So you're gonna go ahead and copy that and paste that into VPS tell terminal thingy.
28:01So this is the home channel. I'm gonna say yes. This is the user ID I want to use for the home channel.
28:05Hit yes. And now we're basically completely set up with Hermes. What do we have available?
28:11So to start off, tool availability, we have vision. We have browser automation. We have image gen.
28:16We have text to speech. We have terminal commands, task planning, and skills. We also have different files.
28:21So here's our settings. Here's our API keys. Here's our data.
28:24Here is how we can edit configuration. So what I'm gonna do is I'm just gonna copy all of this. Right?
28:29I'm gonna take all of this, and I'm gonna go ahead and hit copy. And I'm gonna go back into our Cloud Code project, paste that in, and then just say, alright.
28:36So we just set up this Hermes agent, and here's all the information that it gave us. So make sure you save this so later if I need help with tools or skills or files, you know exactly how to help us do that. So hopefully now you guys are starting to understand the value.
28:49If you're ever getting confused on something with your Hermes or your VPS, you just open this up and say, I'm working on this agent. I need help with x, y, and z. I've actually saved myself a ton of times doing this.
28:58And also when you wanna start thinking about security and maybe getting a firewall on your VPS and locking it down a little bit, which I definitely recommend you should look into, that is something that Cloud Code is gonna help you out with big time. But Hermes can also help you out with that too. I have my Hermes agents doing, like, a nightly sweep of just making sure that things are kind of locked down in the right spots.
29:15But anyways, now you can see it says, do you wanna launch Hermes chat now? I'm gonna say yes. And this pulls up the command line interface for Hermes.
29:23You can see it's gonna load up. It's gonna let us chat with it right here, and it's gonna look kind of similar to if you use Cloud Code in the terminal. There we go.
29:30So we have our Hermes agent. I love this text up here. We can see available tools.
29:34We can see available skills. You can see there's already a ton of them. There's 85 already installed,
29:39and we're able to see our model, our context window, and how long we've been in this session. So let me just make sure that our connection to ChatGPT is working. I'm just gonna say hello, and awesome.
29:47It has given us a response. Now let's go ahead and start our telegram and see if we can get this connected. So this is the bot that I just made.
29:53I'm gonna hit start, and I'm gonna say hello. And if it's working, then we'll see a typing in the top left up here.
29:59But what you'll notice right now is that it's not working. There's no typing. So what I'm gonna do is go back into our Hermes right here and say,
30:06hey. For some reason, the Telegram connection doesn't seem to be working. I just shot the message off and said hello, and I'm not getting anything back.
30:12And look at that. It's already invoking a Hermes agent skill to understand what's going on. It's doing a plan.
30:17It's running some commands right here. So it's basically going to investigate how we make sure to get this Telegram connection all set up. Okay.
30:24So it actually just sent me a message. It says Hermes gateway is back online. If you sent hello, try it again.
30:29So we'll go ahead and try that again. Hello. And now we can see in the top left, it is typing, and it's actually able to respond.
30:35So it said that it found the issue. The gateway was stopped, it started it again. So that is perfect.
30:40Okay. So we are up and running. We can see we now have Hermes right here.
30:44So let's just go ahead and get onboarded a little bit with it. I'm gonna say, hey, mister Hermes.
30:49My name is Nate. I want you to be my ultimate personal AI assistant, so let me know what you need from me and what, you know, features and stuff that you can actually do for me and how you can make my life easier. And once again, that is actually going to be able to transcribe that audio and understand it and then respond to us.
31:07And what I love about using Hermes, especially even if you're in Telegram, you still get to see the visibility of what it's doing. So you can see its skill view. Can see it's adding some user memory.
31:16So it's saying user's name is Nate Herc. So as I'm editing this video, was like, wait a minute. How did this thing know my last name was Herckelman?
31:22All I said in the voice message was, hey. My name's Nate. And I was like, oh, that's scary.
31:27And then I realized Telegram has my full name. So
31:32yeah. And then it goes ahead and responds with a bunch of stuff. So great to meet you.
31:36I've saved your name and that you want me to operate as a serious personal AI assistant. Here's what I can do. I can do admin stuff.
31:41I can do research. I can do coding, automation, files and documents, voice. Here's some stuff I need from you.
31:47So this is where you'll probably just wanna gap to this thing for five to ten minutes. Tell it about your goals. Tell it about what you're working on.
31:53Tell it about your team. Tell it about skills that already exist that you have, and just giving it some more information so it knows you a little bit better. Now what this is gonna start to do is it's gonna start to build its own environment.
32:02Right? It's gonna build those files that we mentioned, and it's going to potentially start to build out the whole infrastructure for skills and other stuff. So the first thing that I think that everyone needs to do when they set up a Hermes agent is they need to connect it to a GitHub repo.
32:13The reason being, if Hermes goes down for some reason and the VPS is corrupted, you still have all of that saved. So you could take that repo and just wake up a new Hermes agent and sync it to that, and then it's like you didn't lose anything. So that's exactly what we're gonna do.
32:26Before I start to give you all of this information about me, I really wanna just make sure that we sync this project to a GitHub repo. So can you just do some research and figure out how that works? I can go ahead and give you whatever you need as far as an API key or
32:40information about my GitHub account so that you can create this, but I want you to set this up as a private repo for me. And so what this is gonna do, it's gonna do research. I'm gonna show you guys how we give it that API key in a safe way,
32:51and it's also going to help us figure out you know, it's gonna probably build a skill around this. And then what we're gonna do is we're gonna turn it into a cron so that every single day at midnight or something, it automatically backs up everything that we did. So every day, if we ever make changes, our repo is committing every single day automatically.
33:06You can see it's viewing different skills. It already has two skills for GitHub repo management and GitHub auth, which is awesome. It's running some terminal stuff.
33:12It's searching through some files. I'll check-in with you guys once we have an action to take. And while it's doing that, let me just show you guys.
33:18So if you've never used GitHub before, don't be intimidated. It is basically just think about it like a like a shared drive, like a OneDrive. It's a place for you to store your projects and your code bases.
33:28And that way, if you wanted to pick up on your laptop or on a different device, you could still access all of those files. So it's really important that we're gonna set up this automated backup. It's completely free to set up.
33:38Go to GitHub and get an account set up. And now we got a response here that says, okay. I checked the environment, and I researched the clean setup path.
33:46Git is installed, but the CLI is not installed. It actually said, okay. We're gonna do the API with a personal access token.
33:53So create a private repo, initialize Git locally, and add a safe dot Git ignore to make sure that our API keys and our secrets don't get pushed to this repo. Even though we're gonna keep it private, still best practice.
34:03It says I need you to send me these four things, your GitHub username, your private repo name, your Git commit identity, and your GitHub token. Before pushing, I'll make sure we do not upload secrets that are in these files, and I will also create a dot Git ignore so it knows GitHub way better than I do, probably better than you do unless you're coming from a GitHub background.
34:22So if you're curious about it, just ask it, and it has these skills to manage GitHub repos. And now what's interesting is if I actually go and give it my personal access token, it should be able to set up the repo. So I don't wanna manually click around and do that.
34:34Let's see if it can do it for us. But what I do need to do is get our token. So I'm gonna click on this link it gave us.
34:39That opens up my GitHub. I'm gonna go ahead and generate a new token. So I'm gonna do a fine grained new token.
34:45I'm going to real quick authenticate. Token name, we're gonna call this the YouTube Hermes.
34:49I'm going to just have this one. Probably you want this to never expire, but because I'm gonna delete this right after the video, I'll just keep it at thirty days. You could do public.
34:57You could do all, or you could do select. And so this is, I guess, the point where you would probably wanna make the repo if you wanted to only sync to that one. So I'm just gonna go ahead and say all repos.
35:05And then for permissions, what we need to do is we want to do contents, and we wanna make sure that this is read and write. So it can pull in stuff, but also push updates every night.
35:15We're gonna go ahead and generate that token, and I am going to now just have to copy this and show you guys how we give it to Hermes. Because what you might be tempted to do is just drop it in the chat. And honestly, like, depending on the model you're using, it might not be a huge deal, but it's just not best practice.
35:30And if you do accidentally do it and it's going to, you know, OpenAI servers, then you can just rotate it. So it's not a big deal. If you're using an open source model and it's completely local and private, then you could just drop it in because Hermes obviously will take it, put it into the dot NETV, and put it where it needs to go, which is which is nice, but then it's in the conversation history.
35:47So the way that we're gonna do this is we're actually gonna go back to the VPS, and then you're gonna go ahead right here and click on open. And what this is gonna do is obviously open up that chat, but we're gonna try to get out of the Hermes chat and just get into, like, this Docker container. So I'll hit control c.
35:59Now what I can do is Hermes config set, all caps, GitHub underscore
36:06token, hit space, and now I can paste in that GitHub token. And when I hit enter, that basically sets that in the slash opt slash data slash dot ENV. So now we have put this token into that dot ENV file inside of our VPS without putting it into the AI conversation window.
36:23And this is the way that you should be setting up all of your API keys in here. So now if I say, alright. So I just dropped in my API key for GitHub
36:32in the dot ENV file. It is called GitHub underscore token.
36:37So see if that works. And then what I want you to do is actually just create the repo for me. My GitHub username is Nate Herk AI.
36:44My commit identity can just be Nate Herk, and you can call this repo whatever you want and make sure it's private. Okay. So hopefully that works, and hopefully it's able to find that API key.
36:54Let's go ahead and just see what it does. Now once again, if this stuff down here is confusing you or if you wanna delete a key or whatever it is, then you will go to your Cloud Code project. You will give it the info it needs and say, hey.
37:05By the way, my agent is set up in a Docker container, and this is the name of the container, and here is the SSH, and help me just figure out where my files are or where my API keys are. Okay.
37:16So that worked. It found the API key, but what happened is we didn't give it permission to actually create repos. Okay.
37:22Cool. So I'm actually gonna just go ahead and create a new classic token. But remember, we can't have two tokens in there called GitHub underscore token or they're gonna clash.
37:29So I'm actually glad this happened because what this will do is it will have me have to show you guys how we can delete an API key. So I'm just gonna say to our Hermes, okay.
37:38Can you give me the command to run inside of the terminal that we can actually open up that dot ENV file that you just accessed? Because I need to delete that GitHub token so I can give you a new one. Alright.
37:50So we have this command, which is the file path for that dot ENV. And if I go to the terminal now, the one up here, which is our root level VPS terminal, not the docker one, and we paste that in, that should basically open up that file for us. Although it looks like that we haven't actually installed Nano yet.
38:06So I'm gonna say when I pasted in that first one, it said nano colon command not found. And I know that I had this, like, pasted in a little bit weird, but I think that we still would've got this either way. So now we can just use this instead apparently.
38:18So let's copy that, and let's get this pasted in here. Hit enter. And now we see.
38:24Can you just show me how we could get the nano command to work instead? And just wanna confirm that we should be doing this on the root of the VPS, not inside of the Docker container image that you run on. Okay.
38:35So that v one command was weird, and I'm normally, I use the nano, so I just asked that question. Okay. So I figured it out.
38:42What happened was the agent's running inside of a container. Right? And then we were doing
38:48looking at the root file for the dot ENV, but, actually, we needed to be looking at the dot ENV inside of that Docker container.
38:55So this is the nano command that we actually need to use that gets us into the Docker container dot ENV file. So now you guys understand that you don't have to understand, like, exactly how all this stuff works. You just have to be able to communicate clearly what's wrong and what you're seeing.
39:10So I'm gonna go ahead and delete this GitHub token, and then we're gonna go ahead and generate the new one, and you guys are gonna see me put it in the same way that we did earlier. What you do in here is you do control o, and then you hit enter, and then you do control x, and that's how you save it. But now I can go ahead and create a classic token instead of a fine grained.
39:27We're gonna do a new classic. And now we can select the scopes. And this is where you would choose exactly what actions you want.
39:32We definitely want the repo actions to be able to access public repos and invitations. And this is where you might wanna just go through and give it, like, read access to things, but not write access to everything.
39:42But as you guys can see, it's not a huge deal if you need to come back in later and increase or decrease the scope. So I'm just gonna go ahead and generate this token. We're gonna copy this, and we're gonna do that exact same thing in here where we do our Hermes
39:54config set GitHub underscore
39:57token, and then we paste that puppy in.
40:00And now that should be set. I'm gonna go back into Hermes and say, awesome.
40:04The new GitHub token that you see in that dot ENV file should be the updated one. So go ahead and create that private repo for us now. And what you see here is even in Telegram, it can ask you for access to things or permissions.
40:16So you can allow it once, you can allow it for the session, or you can always allow something. So in this case, when it is creating a repo or it's going to be, like, committing to a repo, I'm just gonna go ahead and do always allow so that in our skill we're about to set up where it does a daily sync, it just does it with no problem.
40:31Okay. So that has been created. It gave us a link.
40:33If I click on this, it should pull up a private repo right here. You can see this is Hermes personal AI assistant. It's private.
40:39And here are all the files that are already pushed into what we're doing in this project. So if you wanted to dig in, you could take a look at what is here.
40:46Now that that's created, let's go ahead and make a skill. Awesome. So I'm gonna be using you a lot, and you're gonna be creating different skills and different memories about me and stuff like that.
40:55So what we need to do is set up our first cron job, and I want you to build a skill around this. Basically, every single night at 12AM, so midnight central time, I want you to push changes
41:06to this GitHub repo. This is gonna be kind of our, you know, our source of truth. So if you have any questions about that, please ask.
41:12But otherwise, just go ahead and set up the skill and set up the cron. And it's really just as simple as that. In natural language, you say, hey, every night at 12AM,
41:21or you can even do things like what I do for my YouTube videos is I I drop a video, and then I give it the link to that YouTube video, and I say, hey. For the next twelve hours, run a cron every ten minutes to go check on comments and respond to them.
41:35So you can set up basically the same way that, like, the slash loop works in Cloud Code. You can say, hey. For the next twenty four hours, just do this every five minutes.
41:42And then once twenty four hours has passed, go ahead and kill that cron. And it can do things like that as well. You can see it's viewing a bunch of different skills.
41:48It's gonna run some terminal stuff, and then hopefully, we're gonna see it create a skill and maybe update some stuff in its memory. While this is running, there's one thing that I wanted to hit on real quick, which is basically like, what's the difference between using Hermes in the terminal or using it in Telegram?
42:03So functionally, it's really not all that different. It's the same agent in both interfaces.
42:08Telegram does not run a weaker version of it, but you have a little bit less control when you're using it in Telegram. The , and Telegram is more like your remote control. You can also set up things like your dashboard, and Hermes has its own Kanban board so you can monitor, like, tasks, which is in my opinion, I don't use it at all.
42:26I think if I was doing, like, hardcore coding and I had different agents working on different things and I could view, you know, like, the project in a visual way, it would be super helpful. But the way that I'm using Hermes agent, like I talked about in the beginning of the video, where it's kind of like my on the go wherever AI agent, I don't really need that Kanban board.
42:42It's not too useful for me. But anyways, that's like the mental model. The CLI, which I hardly ever use, is best for, like, deep work.
42:50You're building something. You're coding. You're, you know, you're living in there as kind of like your operating system,
42:55and it just has more, like, commands. You you have the ability, obviously, as we saw in here.
43:00Can see your context better. You can manage that better. You have all of the slash commands available to you, whereas in Telegram, it's not exactly the same.
43:07And the session or the context window feels a little bit more ambiguous, and ultimately, it is. It's still doing, like, auto compaction and stuff under the hood, but that's why I'm saying, like, in Telegram, I'm not gonna be vibe coding apps and projects because I might be in context rot territory, and I don't have really the best ability to manage that.
43:24So for me, being able to say, hey. Check on ClickUp and check-in with the team and do this and, you know, do this cron, it's not like a super, super high risk operation, which is why I'm fine having less visibility into context window, session management, things like that.
43:38So that's what Telegram is best for, scheduled things, quick tasks, you know, kind of your general knowledge work that's not super, super high risk. Once again, though, same agent, same brain, same window, same skills, same memory, but you lose a lot of the visibility there. You lose a lot of those slash commands.
43:53And context. Right?
43:55It's token based. It's not message based. So the model, no matter what, is going to see your system prompt, you know, your user dot m d, your soul, all that kind of stuff, and that has to fit inside the context window.
44:07And it's going to be running those auto compactions as you get near that. So in Telegram, it's a little bit tougher to understand, like, okay.
44:14Where did the session actually reset, and how much information does it see in this current working sort of memory? So hopefully that makes sense, the difference between CLI and Telegram, and that's why if you were trying to vibe code like a hardcore app or game out of Telegram, it might just not feel as good as if you were doing it in the CLI.
44:32But anyways, let's see what's going on here. So it set up the first cron job. You can see if we look at this stuff, it was able to write some files.
44:38It was able to use a cron job create tool. It used the cron job list to see if it's there, and it also updated memory at the bottom. So that's super awesome.
44:47The skill is called nightly GitHub sync, and it's syncing to this repo. So the container is running in UTC.
44:53So instead of using a fixed UTC time that would break during daylight savings, I made it run hourly and self checked central time. So this actually syncs to midnight my time in Chicago, and it's going to
45:06mirror safe assistance state into the repo under this branch. And now what else you could do is you could copy this and give this to your Cloud Code project if you wanted it to have this visibility, or, of course, you could have your Cloud Code project look at the repo you're building. However you wanna keep things a little bit synced up, whatever makes you feel more comfortable.
45:22Because, like, you don't have to remember all this. Cloud Code can. Hermes obviously does.
45:26Just give yourself a little bit of, like, insurance there. Anyways, that nightly sync is now active, so that's great. But, alright, I think at this point, you guys really have everything that you need.
45:34You understand the pillars. You understand how to do API keys and, you know, navigate your VPS environment.
45:40At this point, it's really just a matter of figuring out what workflows make sense to put into your Hermes agent. So we wanna start talking about that a little bit.
45:48You have two main paths to have your first skill. The first one is where you describe an outcome. You just saw that what I did with the GitHub sync cron.
45:56That was super, super simple. The other path is you could write your own or you can install one from your Cloud Code projects or obviously from the skills library where you would click on a skill and you would just run this command or tell your Hermes agent to go install that. So for example, if I went ahead and just copied this URL,
46:13put it into here, and said, I want you to go ahead and real quick install the HyperFrame's official skill from here, and then generate me a five second video, which is just like you introducing yourself and showing me a little bit of your personality, if there's anything in your sold out MD already. So while that's running, we'll continue talking about it.
46:30So and then, basically, as you're asking it to do more things and as you're asking it to work on those repeatable processes, watch what happens. Like, I think that's what's really important is to watch what it's doing.
46:40You know, watch if it's viewing a skill. Watch if it's running something in the terminal. Because if you wanted to invoke a skill and it's not, then that's
46:47basically an indicator for you to say, hey. Whenever I say, you know, something along the lines of this, you should probably be invoking that skill. So go ahead and update the YAML front matter so that you more accurately actually call on the correct skill.
46:59And from there, you just use it more and you give it as much feedback as possible, give it more information about you. Now a few things I wanted to touch on about, like, the mindset of having a personal assistant like this. I typically always set up my Hermes agents with their own accounts.
47:12So if I'm gonna give this thing an email address, I'm gonna give it its own Gmail or its own agent mail account. I'm not gonna give it mine. Or if I am, I'm gonna give it an API key with very strict scopes.
47:22I'm also giving all of my different agents different API keys if they're gonna spend. So OpenRouter or Perplexity. I'm gonna give each one a named API key so I can see which agents are using how much of my money.
47:33And I think best practices is like, pretend this is an actual intern or a new employee. What access would you give them?
47:39You wouldn't just give them your credit card. You wouldn't just give them all the stuff. So why would you do that with an autonomous agent?
47:44I'm gonna go ahead and approve this command real quick. So I think it's really important to be thinking about it in that way, obviously, to protect yourself and to protect your business. And speaking of protection, another thing that you wanna look at probably doing is when you come into your VPS,
47:57think about how you can lock this down a little bit more. So you can come over here to security, and you can set up a firewall. And right here, can see I've got one on this VPS, and I think it's overall on my VPS called UPC Guard.
48:08But you can set one up to lock it down for, like, your IP and to block out certain ports and stuff like that. And I don't really know anything about firewalls as far as, like, formal education. I had to do my own research.
48:19But guess how I set up my firewall? I asked Hermes and I asked Cloud Code to do research, look at our environment, look at our VPS, and help me figure out how to lock it down.
48:29And then what you could do is build skills around it where every night or once a week, they're doing an audit on the security. They're maybe trying to attack it, trying to get in, and they're helping you optimize just to make sure that your VPS is staying safe and secure. But anyways, we're starting to reach the end of this video here, so I wanted to wrap up with maintaining your Hermes agent.
48:45Definitely let me know in the comments what else you wanna see with Hermes if I can expand on some stuff deeper for you guys or specific use cases. But anyways, when the agent gets something wrong twice, correct it on the spot and tell it to update the relevant skill and or memory.
48:57So same thing, if you give the same instruction twice, ask Hermes to write a skill for it. When the agent is too verbose or off tone, you have it edit the soul. When you want a new scheduled task, you build a skill and then you just ask it to schedule that cron.
49:08When something breaks, check the memory dot m d. Stale memory is the number one cause of weird agent behavior. And this isn't a tool you finish setting up.
49:15It's . And, of course, you can always at any time say, hey. Read me your memory file.
49:22Read me your soul file. Let me see what's actually in there. And it's funny.
49:25Right here, remember how I asked this agent to make a video about itself to show me its personality? It had to obviously read its own soul file to know what to put inside of that video.
49:37And it's only gonna be a five second video, so we'll see what it really comes out as. But this is just showing you how I dropped in a skill. It's installing it.
49:44It's doing all the hard work, and we should get an output very soon. Alright. Fingers crossed that this is good.
49:50It installed the Hyperframe skill, and it generated a five second video based on its soul. I don't even know it was in its soul. So let's take a look what we got here.
49:56We have Hermes, helpful, knowledgeable, direct, think, build, research, automate. Nate, I'm your action engine.
50:03So for a five second video, not too bad. I turn messy goals into finished actions. Yes.
50:09You absolutely do, Hermes. And I didn't even give this poor thing a name yet. So from here, I'm gonna start building up the soul.
50:14I'm gonna start giving it way more information about me and my business, and I'm gonna keep letting it build up skills and just learn more over time. And that's exactly what you guys should be doing now. Also, one more quick thing is I obviously did this for a demo, so I told it to pause that cron for now.
50:27And I wanted to show you we finally hit that threshold where it does a compaction. So we were at almost a 170,000 tokens, which is over the threshold of about a 136 k. So it decided to go ahead and try to compress.
50:39For some reason, that failed, so it inserted a fallback context marker. And then we can see it can use the other cron tools to list and pause, and then it did a memory update as well. Let's say you didn't understand what these two lines mean.
50:50Once again, just paste that in and say, okay. So you just sent me these messages, I don't understand what that means. Explain it to me.
50:57So I'm not gonna read this out right now because I feel like I've been talking to you guys for a long time. If you give that a quick read, go ahead and pause the video and understand what that means in case you get that message from your Hermes agent at some point.
51:09Now as you start to make more and more of these Hermes agents, it does get interesting. You have some decisions to make, which is basically like where do they live and when do I do that and how do they have separation but still be able to talk to each other. Questions like that start to come up.
51:23So I wanted to talk a little bit more about how you kind of scale this up. So if you're doing it on one VPS, I think the best way to do it is to have them each in their own container, which is why in this setup, I decided to do the one click install besides the fact that it's very, very easy.
51:38So for example, if you guys remember, we have our kind of, like, personal main one that we set up. And remember when we were trying to figure out the API keys and deleting them and stuff, we figured out that those API keys were stored inside of that Docker container. So this agent has its own memory, its own tools, and its own private keys.
51:54So then as we started to add on more, if we put them in their own container as well, they will all have their own keys and then they don't clash. And that's where you can get more visibility as far as like how often they're using their tools and how much money they're costing you and things like that. And that's the value of having, you know, your VPS being like the office building and then each agent being their own container with their own passwords,
52:14you know, keyboard, whatever you you know, if you're thinking like an office analogy, coffee mug, whatever it is. So that ENV file, the dot ENV, first of all, it doesn't get committed to GitHub, so you're not pushing your secrets out there even though it's staying private.
52:27But then each agent's basically having their own, as you can see. If you have a marketing Hermes and you have a finance Hermes, they will not be sharing keys. They will not see each other's keys.
52:36And you can use the least privilege rule, which is basically give each agent only the credentials and the tools needed for its job. So for example, your marketing agent doesn't really need maybe read access into your QuickBooks, but your finance Hermes probably does. And then a quick little decision tree for you guys on when you should create a new Hermes because really for a while, you're probably going to be all set with just the one.
52:57And even if you have one main one that starts doing a little bit of finance stuff, some finance skills, a little bit of marketing skills, what happens is, okay, when you wanna scale up and create a dedicated finance Hermes, you can migrate those tools and those crons and those skills super super easy. And that's the cool part about all of that just living as a markdown file.
53:15So then you take those and you move them over. But anyways, start with the task or the role that you're thinking about making an agent for. Does it need different permissions, secrets, or tools?
53:23If yes, go ahead and start a new agent. If no, keep on moving. Does it need separate longer term memory or separate long term memory?
53:31If yes, create a new agent. If no, ask the next question, which is, is it ongoing repeated work? If yes, create a new agent.
53:37If no, keep it in your main personal because if it's a one off task, you don't need a full new container and a full new Hermes setup. So simple rule, if it needs its own memory, its own tools, its own credentials, its own schedule, or audience, then feel free to split it up into its own Hermes agent.
53:51So I would recommend that you maybe look at, like, a road map of, okay, these are maybe the next one or two or three agents that I'd want. But once again, I would say try to get as much use out of your main personal one to start just because you are still learning how the skills work and how to work with your Hermes and how to set up everything.
54:09So limiting the amount of, like, distractions really and just keeping one, working on that one really well, you'll be able to get pretty far. That's how I would recommend starting out.
54:19And obviously, as you start to get more, you might get confused about like, okay. How do I have this hierarchy between them and how do I make them, you know, talk and assign more to each other? That is something that your main Hermes agent, the one that you're building right now, whether you wanna call that like your COO or your executive assistant or whatever,
54:34that main one is going to be able to help you figure out all that delegation and stuff like that. You just have to have some conversations with it and have it help you plan. But ultimately, there is a point where you probably get to certain scale and you need to, you know, start to segment some stuff off.
54:49So a bad pattern would be one mega agent with all the API keys, with all the skills, with so much bloat of different tools and different crons running, which could cause high confusion and also high risk if something happens to that one agent for some reason. But as you start to scale up, you can adopt a better pattern where you're starting to split up things whether that's via vertical or whether that's via, you know, social media platform or whether that's whatever your separation that makes sense in your business and in your workflows and in your SOPs and in your job descriptions,
55:18that is where you wanna get a little bit more separation. You lower the risk. You, you know, you aren't putting all your eggs in one basket.
55:24As they say, cleaner memory and probably easier debugging and better visibility. So wanted to hit on that real quick, but once again, just because you can spin up a bunch of agents in one VPS doesn't mean you need to. So don't force it.
55:35Just let that phase happen naturally once you've really felt good about Hermes and once you feel like you, you know, have gone through this decision tree and you hit some of these criteria where you need a new one. And then of course, you can keep your Cloud Code project updated and you're gonna be really happy that I told you to set all this stuff up so you can understand the different containers and the different where the files are and all that stuff, this is going to come in handy big time for you.
55:58Now Hermes also does come with, like, its own dashboard where you can look at the recent sessions, you can see your different connected platforms, and it is pretty helpful. But honestly, I don't ever find myself going in here because one of the things that I talked about earlier is the fact that I mainly like to use my Hermes agents when I'm out and about and I'm kind of on the go.
56:16But if I'm sitting down on my computer, I'm usually just working inside of Cloud Code. And this dashboard, it's usually just easier to open up on a local device because you have to open up the tunnel and there's a gateway, and you guys will understand once you start setting it up.
56:28But if you wanna check it out, this is where you can also have your Kanban board, and you can also look at things like, you know, different keys and different configs and skills and plugins, and it's just basically a nice little visual dashboard to see what's going on here. You can also look at your crons and you can create new ones from here.
56:43So what you do is you'd go to your Hermes agent and you'd say, hey. I wanna open up the Hermes dashboard. Can you help me figure it out?
56:49What you're gonna need to do is you're gonna need to say, okay. Here is, like, my VPS route, and here is my Docker container setup, if you have that set up like that.
56:57Because you have to kind of, like, open up the gateway and make sure there's a tunnel open so that you can open up this local dashboard, and that's where you can start to play around with this.
57:06So the first time you do it, you might feel a little bit like this sucks because the first time you might be pasting different things around and it might not feel like it's gonna work, but it will work. Just keep giving
57:17CTAyour Hermes agent the info, keep giving it what you're doing, and it will get you there. And once you get there, then say okay, save this to your memory. Turn this into a skill so that every time I ask for the dashboard,
57:27CTAyou give me the three commands I need to run, and then I just boom, boom, boom, run them, and I'm good. So I just wanted to warn you guys about that. And, you know, if you wanna get in here and play around, the Camber board is pretty cool.
57:36CTAIf you got multiple agents running, you can sort of like assign them to different agents, and they'll pick up tasks, and you can see them move around. But like I said, I didn't really spend too much time on this in this video because I don't ever find myself in this dashboard, but it is a nice built in feature.
57:50CTASo anyways, don't forget, I'm gonna put all of this information that we talked about today and the full setup guide and all that stuff into a free resource guide that you can access in my free school community. The link is in the description. Don't forget to use the link to go to Hostinger and use code Nate Herc to get 10% off your annual plan when you set up that VPS.
58:05CTAAnd, yeah, let me know what else you guys wanna see with Hermes and where I can expand on some stuff. I'd love to bring you guys some more content around what you wanna see. So gonna do it for today.
58:13CTAIf you enjoyed the video or you learned something new, please give it a like. It helps me out a ton. And as always, I appreciate you guys making it to the end of the video, and I will see you in the next one.
— full transcript
§ 05 · For Joe
Your agent is only as good as its context files.
Steal this system
Nate's Claude Code VPS-agents project — one repo, one .env per agent, all passwords and IPs in one place — is the missing organizational layer most builders skip on day one.
- Create a Claude Code project called vps-agents right now. One subfolder per agent. .env with IP, root password, admin creds, API keys.
- Five pillars first, tools second. Before installing anything, write user.md and memory.md. The agent is useless without durable context.
- Pass API keys via 'hermes config set KEY value' in the VPS terminal, never in the chat window.
- First cron = nightly GitHub backup. Natural language request, Hermes builds the skill automatically. Do this on session one.
- CLI for building, Telegram for running. Use Claude Code to set up and debug; use Telegram when you're walking.
- One agent, one job. Start with your personal/manager Hermes. Spin off specialist agents only when you hit the decision tree criteria.
- Correction + memory = the actual training loop. When the agent gets something wrong twice, correct it on the spot and tell it to update the skill or memory.
§ 05 · For You
How to get more done without being at your desk.
If you want an AI that works while you walk
The most useful thing Nate demonstrates isn't the install — it's the mental model: your AI assistant wakes up with no memory every session, so the files you give it are everything.
- Think of your AI assistant like the guy from Memento. It forgets everything overnight. Your job is to write the sticky notes it wakes up to.
- Voice-message your agent from your phone while you're walking. It transcribes, thinks, and replies — no keyboard needed.
- Set up one cron job on day one: nightly backup to GitHub. Even if your server breaks, you keep all your agent's memory and skills.
- Don't give your AI assistant your master password or main accounts. Treat it like a new hire — narrow access, build trust over time.
- The more you correct it and ask it to remember things, the better it gets. The training is the usage — there's no separate setup phase.
