The argument in one line.
Hermes is a superior AI agent harness to OpenClaw because it enforces disciplined memory management, automatically builds its own skills through use, and maintains stability through a philosophy of simplicity rather than feature bloat.
Read if. Skip if.
READ IF YOU ARE…
- You run OpenClaw agents in production today and spend recurring time debugging or fixing broken workflows.
- A hobbyist or small team building AI agents for personal projects or internal tools who wants a system that improves autonomously over time.
- You're comfortable with cloud VPS setup and Python installation, and you want a walkthrough from install through working Telegram bot integration.
SKIP IF…
- You've never used an AI agent framework before — this assumes you understand agent concepts and compares Hermes specifically to OpenClaw, not foundational agent theory.
- You need production-grade enterprise support, compliance documentation, or SLA guarantees — this covers open-source community tool setup, not commercial deployment.
- You're evaluating agents purely on benchmark metrics or academic performance — this breakdown prioritizes practical stability and developer experience over raw capability scores.
The full version, fast.
NetworkChuck argues that Hermes, the new agent harness from Nous Research, is the right replacement for OpenClaw because it grows more useful over time instead of degrading into a brittle project you constantly maintain. The mechanism is a disciplined memory system with hard character caps on the user and memory files, background nudges every ten turns that force the agent to curate what it knows about you, plus a self-improvement loop where the agent writes and refines its own skills as it solves real tasks, with a curator pruning stale ones. The practical implication for you is to install Hermes on a cheap VPS, wire it to Telegram and your preferred model, seed it with identity and environment context, then let it build its own capabilities through use rather than hunting a skills marketplace.
Members feature
Chat with this breakdown.
Modern Creator members can chat with any breakdown — ask for the hook, quote a framework, find the exact transcript moment. Unlocks at T2: refer 3 friends + add your own API key.
Create a free account →Where the time goes.
00:00 – 02:22
01 · Why I switched — the 5-reason teaser
Chuck opens with the hook (agent that grows with you), explains tool fatigue, previews all 5 reasons and the IT Hogwarts wizard demo, and teases the cofounder interview.
02:22 – 03:36
02 · 1. The Vibes
Hermes website aesthetic and Nous Research mission (humanistic, censorship-free, democratic AI) sold Chuck before the product did. Cofounder Jeff describes origin: Discord hackers building open-source AI.
03:36 – 05:36
03 · Install — Hostinger VPS (sponsored)
One-line install on a $5 Hostinger VPS. Coupon code, Ubuntu setup, SSH in, paste the Hermes install command. Hostinger is the demo vehicle, not a bolt-on sponsor.
05:36 – 09:46
04 · Setup — model + Telegram
Choose inference provider (OpenAI Codex using existing ChatGPT sub, Grok, OpenRouter, local LM Studio). Telegram bot via BotFather. Ron Weasley persona seeded. User ID whitelist security.
09:46 – 15:00
05 · 2. Memory — how Hermes actually learns
Hard file limits: USER.md (1375 chars) + MEMORY.md (2200 chars) force curated distillation not bloat. 10-turn background nudge updates files mid-session. Live demo: seeding Ron persona, watching SOUL.md, USER.md, MEMORY.md populate.
15:00 – 17:01
06 · Honcho — long-term peer card memory
Optional Honcho service reasons across all past sessions, building a peer card. Chuck used it in Tokyo. Honcho surfaces uncomfortable truths about Chuck himself (high-friction technical procrastinator). Works with OpenClaw too but is first-class in Hermes.
17:01 – 19:00
07 · 3. The people — Hermes existed before OpenClaw
Nous Research built Hermes 6-7 months before OpenClaw as internal recursive self-improvement tooling. When OpenClaw launched they compared and found theirs less clunky. Jeff Quesnelle cofounder clips throughout.
19:00 – 24:10
08 · 4. Self-improving skills — the headline feature
Hermes auto-generates reusable skill files from completed tasks. Live: Ron installs Twingate headless client, then creates twingate-client-operations skill unprompted. Curator agent runs in background pruning stale skills (active/stale/archive states). OpenClaw = skill marketplace. Hermes = skills crystallized from your own workflows.
24:10 – 27:15
09 · Live demo — Home Assistant + UniFi
Ron receives Home Assistant IP and API key via Telegram. Turns off Chuck lamp, changes color to blue, closes automatic blinds (filmed live). Given UniFi API key, creates UniFi network operations skill. Two parallel sessions running simultaneously.
27:15 – 29:44
10 · 5. It just does not break
Month of use, zero unexplained failures. Wife uses it daily for homeschooling and household management (6 kids). Hermes = product. OpenClaw = project. Nous Research team prioritizes depth over feature bloat.
29:44 – 30:57
11 · Dashboard, Kanban, computer use
New Hermes dashboard: skills and plugins, multi-agent profiles, auxiliary models, achievements. Kanban task board for async agent work (hit rate limit live, shown unedited). Computer use in preview.
30:57 – 32:39
12 · Outro — prayer
Chuck signature prayer for the audience. Authentic brand ritual. The mechanism: personal end-of-video ritual creates community identity and viewer loyalty.
Lines worth screenshotting.
- Hermes topped OpenClaw on OpenRouter token usage and became the fastest-growing AI agent GitHub project — the market voted with its tokens.
- An agent you trust enough to give to your wife is a different product category than an agent that requires constant maintenance — reliability is the actual feature.
- The self-improvement loop in Hermes means it writes its own skills and gets better on day 30 than it was on day one — most agents stay static.
- AI tool fatigue is real: adding another harness that breaks creates more work than it saves, which is exactly why an agent that doesn't break stands out.
- Hermes has a built-in OpenClaw migration path — the product is explicitly designed for people who are leaving the current market leader.
- An agent that learns from your usage and accumulates skills over time is a compounding asset; an agent you have to reset every session is a consumable.
- The vibe and mission of a tool are not soft factors — they signal whether the team has long-term conviction or is shipping for the trend.
- Giving an AI agent full control of smart home devices and network hardware requires a level of reliability that most agents don't meet — Hermes was the first one trusted enough to try.
- Bounded memory — knowing what to remember and what to forget — is smarter than storing everything, because an agent with too much context becomes as confused as one with too little.
The skill that writes its own skills.
Agent architecture playbook
Hermes wins not because it does more things, but because it remembers the right things and crystallizes your workflows into reusable skills — so day 30 beats day 1 instead of degrading.
- Use bounded memory limits in any agent context you build. Hard caps force distillation and prevent system prompt bloat — this applies directly to CLAUDE.md design.
- The product vs project frame is a clean positioning knife. Apply it to JoeFlow vs competitors: does it feel like something you use or something you maintain?
- Chuck's install-during-the-video format is worth copying for JoeFlow content. 'You will have X running by the end of this video' is a promise that earns the watch.
- Self-improving skill files are essentially a CLAUDE.md that the agent writes and curates itself. Worth exploring for JoeFlow session workflows.
- The cofounder interview splice-in is a trust multiplier. If you interview anyone about their tool, cut the best 30-second clips into your review — it is authority plus Academy upsell in one structural move.
Terms worth knowing.
- Hermes (AI agent)
- An open-source AI agent framework that installs locally or on a VPS, learns from interactions over time, writes its own skills, and is designed to improve the longer it is used — positioning itself as a self-improving personal AI assistant.
- OpenClaw
- An AI agent platform and one of the earlier widely-used autonomous agent environments, which Hermes is compared against as an alternative that requires less maintenance and fewer breakages.
- Bounded memory
- A memory architecture that stores only the most relevant and recent context rather than an unlimited history, preventing the agent's memory from growing unwieldy and keeping retrieval fast and accurate.
- VPS (Virtual Private Server)
- A rented virtual machine hosted in a data center, used to run AI agents and services continuously around the clock without depending on a personal computer being on.
- Telegram bot
- An automated account within the Telegram messaging app that can receive messages and commands, used here as an interface for sending instructions to and receiving responses from a self-hosted AI agent.
- Self-improving agent
- An AI agent that writes and saves new skills based on tasks it completes, so its capabilities expand automatically over time as it encounters new types of work.
- OpenRouter token usage
- A public metric tracked by OpenRouter showing how many API tokens each AI model or agent framework is consuming across all users — used as a proxy for adoption and growth.
- Smart home integration
- Connecting an AI agent to home automation devices — lights, thermostats, plugs — so the agent can monitor and control them through natural-language commands.
Things they pointed at.
Lines you could clip.
27:55
“Hermes feels more like a product. OpenClaw feels like a project.”
16:46
“High friction technical procrastination — gravitates towards tool building, wiring, to avoid high-stakes communication or soul work.”
20:46
“We struggled through things. When we figure out ways that solve hard problems, we note that down, and then we iterate on those successes.”
00:20
“I gave it to my wife. She calls hers Honey. It's her BFF.”
Word for word.
metaphoranalogy
00:00I'm switching to Hermes. The vibe, the mission, that alone sold me. But the idea that the Hermes agent grows with you, it's gonna be better on day 30 than day one.
00:09That got me hooked. Also because I'm tired of fixing my OpenClaw agents, and I'm not the only one. This is now the fastest growing GitHub project.
00:16It just topped OpenClaw on the open router token usage. This thing's taken the world by storm. It can't be that good, can it?
00:22Is it actually worth diving into yet another tool, another harness? Let's talk about it. I've been using it for a month, and it's the only agent I felt comfortable enough with to give to my wife.
00:31She calls hers honey. It's her BFF. I'm telling you this thing is different.
00:34Let me show you the five things, the five reasons that made me switch all my stuff from Open Claw to Hermes. Actually, here's a teaser. The vibe, the memory, the fact that it learns.
00:43Also, we're gonna build our very own IT Hogwarts themed troubleshooting wizard. Easy for me to say. It's literally magic.
00:49And also went a bit too deep, I think. I ended up talking to one of the cofounders, Jeffrey Cannell. So you're gonna hear from him too.
00:56Now I'm trying so hard to not use the phrase right now because I told you I would chill out on that this year. But this, it's something special, and I don't want you burning tokens on tools that will frustrate you.
01:06This one's worth your time. So get your coffee ready. It's time to enter our self improvement loop.
01:14Self improvement loop. What does that mean? That's number four.
01:16It has to do with a skill system, and it's probably the best reason on this list. But hold on. Don't go there yet.
01:21You haven't installed it yet, and that's my goal with this video. You're gonna walk away with Hermes running, and it's actually pretty easy. They even have an open claw migration path.
01:28How nice of them. But, hey, I get it. You might be where I was about a month ago.
01:33AI tool fatigue. Don't give me another tool. I'm full.
01:36Get it away from me. But Noose Research, the company behind Hermes, told me to try it, and I said, fine. But let me hit you with number one real quick.
01:43Look at this site. How can you not try this for the vibes alone? We're gonna install this here in a moment.
01:47Don't worry. But just sit here and appreciate this. This isn't some random lobster project.
01:51Wee.
01:53This is a company with a mission, a purpose. I asked Jeff, one of the cofounders, what's up with this? As much as we're online, you and me are, like, terminally online.
02:00We're also, like, creatures of this world, and our our sight and our taste and all of that are part of who we are. We put we spend a lot of time to make sure that we have that that that, uh, that vibe feel behind it. And it's not just the vibe.
02:12It's who these people are. That that was our our origin story. It was just a bunch of kinda hackers on a Discord,
02:18uh, trying to figure out if we can make open source AI. We'll hit that here in a moment, but first, we gotta install this. To install it, we just run this one command.
02:25But where? You gotta put it somewhere. Thankfully, you can pretty much run this anywhere.
02:28It's lightweight, built in Python, but my favorite place to run this sucker is in the cloud. Hostinger. That's where I put my wife's agent.
02:34They are the sponsor of this video and the sponsor of getting you spun up on Hermes as quickly as possible. Are you ready? Quick sip of coffee.
02:42Go out to hostingyour.com/network check Hermes. And what we're about to do here is install Hermes on a computer and the cloud.
02:48But, again, you can install it anywhere. I think it's actually gonna be Windows native here soon. And don't worry.
02:52The tutorial I'm showing you right now will apply for anything apart from the hosting your stuff. But here we are in hosting here. We could do the quick deploy option if you wanna be really fast, but I prefer this method.
03:00Don't click on that. Instead, go up here to services. Click on VPS hosting right here.
03:04By the way, a VPS is a virtual private server, fancy talk for computer somewhere else, but it's always on and always available and then always awesome. We'll choose our plan. KVM two is the choice.
03:14It's a whole home lab in the cloud. Run Hermes, run OpenClaw, run Docker. I don't care.
03:18Do whatever you want. It can handle it. And, yes, you can run Hermes and OpenClaw side by side.
03:22Test them. Once you choose your plan and watch this, put in the coupon code network Chuck and watch magic happen.
03:32Boom. It worked. Choose Ubuntu as your OS.
03:34Go through a few more steps and then wait for your VPS to brew. You should land on the VPS dashboard, and we're gonna copy this command right here. Copy.
03:42And then we'll launch our terminal. Whatever OS you have, go to your search bar, type in terminal, launch that sucker.
03:47Oh, yeah. Best place to be. And that's actually one of the things I love about Hermes.
03:51This is a bonus one is they actually do make it fun in the terminal, not just the messaging app, and they do have that too. We'll paste our command, type yes to accept all fingerprints, and then put in the root password that you set up earlier.
04:03And we're in. Right now we are remoted into or logged in to a computer in the cloud, and this is where we're going to install Hermes. And, again, this is gonna be pretty Let's go out to Hermes and grab that one line command I talked about.
04:13There it is. Click on copy. Get back to our terminal, paste that in, and go.
04:19It's gonna do its thing. Coffee break.
04:24This is my favorite part of the videos, honestly. Watching stuff scroll across the terminal while sipping coffee.
04:32It's therapy. Hey. While you're waiting, have you hacked the YouTube algorithm today?
04:36Let's make sure you do. Hit that like button, subscribe, notification bell, comment. You gotta hack YouTube today.
04:41Ethically, of course. Now what we're about to do when this finishes up is we're gonna create our IT agent. And this is something actually I got inspired by from Jeff over at newest research.
04:50They get a dog food mentality over there. They use their tools. And I asked him, like, how do you use it?
04:54He's like, this example. So we actually have our agents in our Discord channels
04:59and have them running, and we talk to them like they're team members, basically. And there's just a Hermes agent who's like the the the develop you know, the the system admin. So someone comes into our Discord and says, oh, something this happened.
05:12I got this error message. We're able to just talk to the Hermes agent. And what's really amazing and is that through having just done this, that agent has built up a huge skill set of skills particularly
05:25related to debugging our infrastructure. When he told me that, my brain exploded, and I said that's what I'm doing. We'll talk more about their example later.
05:31I'll show you some stuff.
05:36No. The vibe. Come on.
05:38That was cool. Okay. We're here.
05:39Time for quick setup. Hit enter. The first step is choosing our inference model or the brain.
05:43Similar to Open Claw, Hermes is a harness. It's a set of tools that would use whatever LLM you want. If you wanna go local with LM Studio, do it.
05:51Hermes is really good with it, especially the Quinn model. Wanna go frontier? Open router's good, but my favorite one to start with is the OpenAI Codex.
05:58Why? Well, because you can use your chat GBT subscription with this. Most of us already pay for chat GBT.
06:03And without paying any more money, you can use chat GBT as the brain for Hermes. Hey, Chuck from the future here. Something epic just happened.
06:10This Grok partnered with Hermes, and you can use this as a user Grok subscription with Hermes. Just another option, and it's pretty powerful.
06:18Let's try it out. Hit enter. And all we gotta do is copy this URL, paste that into our browser, get logged into our chat GBT account or OpenAI account, and then grab this code that Hermes gave us and paste that in here.
06:30Click continue. And that was it. Pretty hard.
06:32Login successful. And we'll choose our model. Go five dot five.
06:36It's the smartest. Select your terminal back end. Keep it local for now.
06:39We'll just know you can run it remotely. That's a whole thing. Then finally, set up messaging.
06:43Let's go ahead and do Here you have options. We're gonna go with Telegram because it's the easiest, but you can do any one of these.
06:49Also noting that it has less options than you would see with OpenClaw. Why? You'll start to see this develop as kind of the mentality behind news research and Hermes.
06:56Instead of supporting everything, they'd rather make the experience great with a few options. Now, obviously, this is more than a few. Let's go Telegram.
07:03It's space to select, hit enter, and we're almost there. This part actually is really easy. It's asking you for a Telegram bot token and to talk to somebody named the bot father.
07:12If you're not already a telegram user, become one and open up telegram. Search for the bot father. He's gonna help us create a new bot.
07:19Open. Click on create a new bot, and let's name our bot.
07:22Mine's gonna be Ron Weasley. That's my IT wizard. By the way, you're gonna watch me walk through every step of making Ron an IT wizard in my company.
07:29Legit. Give him a username, must be unique, and it must end in bot.
07:34Nailed it. Create bot. That was it.
07:37Now that little pixie dust thing, we're gonna copy that. Click on copy. That's your token.
07:42Paste that right here in the terminal. Hit enter, and we're almost done. This is a security measure where you can tell your Hermes agent, hey.
07:48Only talk to me. Don't talk to anybody else. Ignore everyone else.
07:51I'm the user now. Sorry. That was lame.
07:53We're gonna go with it. We're gonna talk to one more bot before we can talk to our bot. The user info bot, we need our username from telegram.
07:59So we'll find the user info bot. And usually when you just start talking to him, you'll get your user ID just like this. Copy that.
08:06Paste it. If you wanna add more, you can. For example, my wife's agent, Honey, I can talk to her as well, not just my wife.
08:12We'll make our user ID the home. Just hit yes. We'll install the gateways, a system d service.
08:16If you're like, what is system d? I got a video on that right here. Just hit yes for how the gateway service should run-in the background.
08:22If you're on a laptop, go with the default option. For VPS, let's choose the system service. So I'm gonna go with don't get mad at me in the comments.
08:30Put run it as which user. I'm gonna put root. Why?
08:34I'm on a cloud machine. That's the only user I have right now. It's who I am right now.
08:37And this machine is gonna belong to Ron. I want him to own it. He is the IT admin anyway.
08:43This is not best practice or advice for any other setup. For our example, it's fine. Start the service now.
08:49Yep. And we're kinda done. And at this point, we could talk to Hermes right here in the terminal or via telegram.
08:55Let's do it in the terminal. It's cooler. Hit enter, and Hermes is here.
08:59Let's see if he is here. Hi. He's alive.
09:04Also notice at the top here, all the available skills that are default built in. Talk about that here in a moment at number four, foreshadowing. Foreshadowing accident.
09:14Let's try telegram and make sure he also is alive there. We'll talk to our new user. Hi.
09:19Typing. Hi, Chuck. What can I help you with today?
09:22Now at this point, if you've never used an agent before, this is magic. Right? Enjoy.
09:27Go crazy. But if this isn't your first agent, if you've already used OpenCall, you're probably going, Chuck, uh, how's this different? This feels the exact same.
09:34And, honestly, at first glance, it kind of is. But let me hit you with number two. I discovered this while walking the streets of Tokyo, and it's kind of insane.
09:41Transition.
09:46I think memory is the reason people switch to Hermes and stay, and they may not even realize that's the reason. Because under the hood, Hermes and OpenClaw kinda do memory in the same way with a few philosophical differences, and they make all the differences in the world difference in the world. Let me show you by building out our IT agent right now.
10:02So remember, this is gonna be wrong. Now if I say right now, who are you? It's gonna be generic.
10:07I'm Hermes. Let's change that. I've got a nice prompt.
10:10I'm gonna feed him with, seed him with. Here is who you are going to be. Update any file, including your soul to reflect this going forward.
10:19And then here's the massive prompt. Essentially, he's Ron Weasley. He's following in his father's footsteps of falling in love with Muggle technology, specifically IT.
10:26Yada yada yada. It's a backstory you can see in the description below. Let's see what happens.
10:31Now notice it's already using a skill. This is actually something I love about it. It shows you the tool use happening.
10:36Okay. Identity locked. Let's test it.
10:38I'm going to do new, start a new session here. We will approve that, and I'll say, who are you?
10:45Nailed it. This is gonna be fun. Okay.
10:48So Hermes edited his soul dot m d file. Big deal. Open clock can do that.
10:52If we cat that, here it is. Still nothing amazing yet, but hold on. Let's start building him out a bit more.
10:58I'll tell him who I am. I'm never checking. Call me Chuck.
11:01They'll also talk to other members of my team when they have issues. Blah blah blah. Let's tell them that.
11:05Now watch what happens here. I think it's gonna happen here. I'm doing this live with you.
11:09There. There it is. Put it in the pensive.
11:12Notice we have a memory action here, and it added it to my user file. This file let's go open it real quick. We'll c d into our dot Hermes directory and then jump into our memories directory.
11:24Well, l s there, and there it is. User dot m d. Let's cat that.
11:28This file is what Hermes will write about you, what it learns about you, and it curates this on its own. And that's a key part. We're we're gonna cover why that's important here in a moment, but let's keep adding.
11:39This is fun. I'll give him some more information. I'll tell him about the network and a few other things.
11:44With that information, let's see what happens. So we add more to the user side. Let me tell him more about him.
11:53Ah, and there's the other part of the memory. He added it to his memory file. Let's go look at that.
11:58LS. There's a new file there. Memory dot MD.
12:01Let's cap that. Notice the memory file is more about the environment where Ron is running some technical things. This will also be curated by the agent.
12:10Now those two things, the user dot MD file and the memory dot MD file, they're not groundbreaking things. Like, OpenClaw has these two. In fact, when you move from OpenClaw to Hermes, that's part of the migration is just taking those files over to Hermes.
12:23The difference is how Hermes uses these files. Now both Hermes and OpenClaw will take these files. And when you start a new session with Hermes, boom, new session.
12:32The user and memory file will be loaded into the system prompt so that when you talk to your agent, it always has the most relevant information about you and what you're doing. Both OpenClaw and Hermes do that, but here's what Hermes does. That OpenClaw does not do well, and you'll feel this.
12:46And this is why OpenClaw feels clunky and bloated on day 30 versus Hermes. Actually, Hermes does two things. The first thing it does is it has hard limits on the size of those files.
12:56The user file can only be thirteen seventy five characters. The memory file, 2,200 characters.
13:02Now why does this matter? Well, it forces the agent to be very particular about what's gonna be loaded into its system prompt. Now, for example, we were just talking with Ron, and he was adding stuff to this memory.
13:12And as we keep talking to him and adding more things and talking and talking and talking, that's gonna get full. So what happens when it gets full? Well, he has to delete something.
13:20He has to curate it, guard what actually needs to go in there, what needs to go in there. And that keeps the agent focused.
13:26It forces the agent to distill what is actually important about interacting with you and what you need. And that actually matters a lot versus open claw, which does have similar mechanisms, but mostly it's gonna end up bloating itself over time if you don't really watch it. This does it by default.
13:41It does it without you even thinking about it, and it did it when I didn't even realize it was doing it. I got a story for you on that here in a moment. This is my wow moment with Hermes.
13:48The second thing it does is it nudges by default every 10 turns. I believe it's every 10 turns.
13:54Have your Hermes agent fact check me. It will run a background agent to see if anything that we've been talking about should update the memory or user file. Most agent infrastructures, including OpenClaw, only do this when you're about to compact or start a new session.
14:09This happens during the session with Hermes. It's more active. This alone made it feel pretty different, and it guards it over time.
14:16But then I tried something else. Watch this. This this is crazy.
14:20I tried this. We'll use the Hermes command line options. We'll do Hermes memory setup.
14:26Here, we can configure the Hermes memory. And right now, by default, we're just using built in memory and user. It also has a mechanism to search your past sessions if it needs to.
14:35But then we have these add ons that do something kind of insane. I'm not gonna dive too deep on this because that could be its own video. In fact, I'm probably gonna make a video about it, but I added this thing called honcho.
14:44You can run a honcho server locally or run it in the cloud. I tried it in the cloud. So here's how this works, and this is crazy.
14:51I talked to Hermes. My agent's name is James, James Potter. Every time I send a message, that message is also sent to Honcho.
14:58Honcho is a peer service. It's not Hermes. It's kind of a plug in that will start to reason over what I'm saying, and it will start to build out what's called a peer card.
15:06Basically, who's Chuck? And it gets to know me. Over time, as more and more messages are sent, it will start to make more conclusions and learn.
15:14But that's not the cool part. The cool part is still back at James. With Honcho working in the background, when I send a message to James and his system prompt, he's getting his user dot MD file, his system dot MD file.
15:24Now keep in mind, these are all very important because when you're talking to whatever model it is, anthropic, GPT, it's gonna start from scratch. It's a blank slate unless you fill its head with something when you start talking to it.
15:34So it has this built in stuff that we give it. But then also Honcho is configured to go, what does James need to know about Chuck right now in this moment?
15:43What context would be important based on what Chuck just asked? This combo right here, I can verify because I used Hermes for a month with this setup. It's kind of crazy.
15:52Let me show you what Hanjo looks like actually on the cloud. Here are some conclusions it made about me, like my daily habits or that Kathy does not like spice. And it builds out this peer card, which I can't show you completely of my personality and my traits and my preferences like this one.
16:09Trait, high friction, technical procrastination gravitates towards tool building wiring to avoid high stakes communication or soul work. Ouch.
16:18So I was in Tokyo when I first started using James. I would get up early before everyone walk the neighborhoods, and just have conversations. It's good.
16:25I had some breakthroughs. Now to be fair, honcho is something you can actually plug into Open Claw as well, but it's not as good. It's not as much of a first class citizen as it is with Hermes.
16:34But, honestly, Hermes is pretty stinking great even without this addition. All this is kinda scary a little bit because you're relying a lot on the AI model to learn things and self correct itself.
16:46And should we allow it to do that? I asked Jeff about this. The models are smart.
16:50Like, like, get out of the way of the models, basically. You know? Like, they're smart enough if we let them to just be figure out what it is that you want to do.
17:01So at this point, we have our IT agent, Ron. He knows who he is. He's ready to start working.
17:06We're about to give him some skills. And, actually, no, we're not. You'll you'll see.
17:10The way it does, this is so cool. But before we get there, did you know that Hermes was around before OpenClaw?
17:18You didn't know that? Me either. I was talking with Jeff, the cofounder, and he told me this.
17:22So actually Hermes agent started out as an internal tool that we wrote
17:26started it almost six to seven months ago as a tool we were using internally to, like, prototype this recursive self improvement for model training. And when Open Claw came out, it was actually kind of weird because, like, we actually already have this. At the time, we were aware of the effect it would have on the world.
17:44You know, when we saw what what OpenCloud did, we're like, we can give this out because we have our own version. And, honestly, when we tried OpenCloud, it I mean, I I hadn't used it before, but some of the other people on Tmall, and they're like, this felt a little clunkier compared to what we had internally. I love that.
17:58They tried Open Claw and was like, wow. This is way clunkier than what we have. Let's release our version to the world because we think it's better.
18:05And this is number three, by the way, of the reason I switched from OpenCLO to Hermes. It's the people and the story behind this tool. We are a group of researchers who found each other because we care about making humanistic,
18:17censorship free, and democratic AI.
18:20And I think that's important. Right? Especially now in the day and age of AI, I think who's making the tools, why they're making the tools, and what they're gonna do going forward is really important for us.
18:29And knowing that Hermes was around before, it wasn't just a reaction to OpenClaw and that it was built by AI researchers, people who are building their own models, which I don't know if you knew that. Noose research, they literally were just a bunch of hacker nerds in a Discord training their own models, which was also named Hermes.
18:46They still have their Hermes models. Hermes agent is separate. But after talking with Jeff, I was even more convinced that this company is pretty cool, and I'm very happy with my decision to switch to Hermes.
18:55AI is not meant to replace you. It's meant to make you be a better version of you every day. And by the way, a lot of what we said, I can't include in this video.
19:04It's just it'd be too long. So if you wanna see that full interview, I'm actually gonna launch it first on the network Chuck Academy alongside a course we're building for Hermes. I'm showing you the basics here to get you up and running and going crazy.
19:14There's a lot more you can do. So check it out. Link below.
19:17Okay. Now on to number four. The most powerful thing that Hermes does, it's the headline.
19:22It's the grow with you thing that everyone talks about. Better on day 30 than day one. Watch this.
19:28We're gonna give Ron some tools, man. We're gonna make Ron powerful. We're gonna give him some skills, but Hermes does skills in the weirdest way.
19:35Watch. It's gonna happen. Right now, he's in the hosting your cloud.
19:38Right? He can't talk to my IT stuff here in my studio. Not yet.
19:42We're gonna use TwinGate. It's a VPN. It's a way to remote into my stuff, and it's amazing.
19:46I'll tell Ron, hey. I need you to set up the headless Twingate client for access to my studio.
19:53Here is your key. Make it happen. Now I'm doing this live with you.
19:58I'm hoping he'll do what I think he's gonna do here. Fingers crossed. He's gonna start working.
20:02He's gonna start building this, and our video editors are gonna blur out my key. Done. Notice he added it to his memory.
20:08Okay. Cool. Now he's connected to my network.
20:11That's cool by itself. Right? Search the network.
20:13See what you can find. Oh, did you see that? There it is.
20:16That's what I've been waiting for right here. I didn't tell him to do this. Self improvement review skill, Twingate client operations created.
20:26Ron just made his own skill. That's the power of Hermes.
20:32It makes its own skills. The skill system, which is sort of like the heart of it, which is the ability for it to crystallize once it's viewed how you operate to take learnings, um, basically, and and crystallize them down into a a, like, a a a meaningful chunk that it can then reuse.
20:48So we sort of modeled it after kind of how, you know, a crude version of how potentially, you know, we ourselves work, which is we struggle through things. When we figure out ways that solve hard problems, we, uh, we note that down, and then we iterate on those successes.
21:04What?
21:05How powerful is that? And that's a huge differentiator between Open Claw and Hermes.
21:10With OpenClaw, it's all about finding skills in a marketplace, whereas Hermes is all about building skills, repeatable workflows based on your interactions with Hermes.
21:21Now Hermes does have a built in skill library that is curated by the Hermes team. One of the, um, prepackaged skills that we do ship is this, uh, GitHub PR review, um, that is is in there. That skill actually is like the result of Technium having gone through thousands of PRs manually
21:41and built up the review process that we use internally for PR reviews. And now it's like a very high quality crystallized PR system. And that's important because, uh, I don't know if you remember.
21:50Open Claw when it first came out? Dude, malware central. Claude hub where you could download a bunch of Open Claw skills that the community was uploading.
21:58Yeah. There was bad stuff in there. Um, Open Claw had a bunch of CVEs or vulnerabilities that would get you hacked.
22:05As of this video, Hermes hasn't had anything agent related hit it yet. And I think it's because of this mentality right here. Also, it's kind of the Hermes team philosophy, keeping things very simple, but it's not just the building of skills that's pretty cool.
22:17It's the self improvement loop that I alluded to at the beginning of the video. Hermes just came out with this, a thing called the curator, which it will be an agent that runs in the background periodically, and it will review the current skills.
22:28It will curate skills. It'll make sure your skills are good. If there's a skill that's not relevant anymore, it'll remove it.
22:34It'll improve skills as you're working with your agent. It actually moves skills through an active stale and archive state, and it exists so that skills created via the self improvement loop don't pile up forever.
22:46And by the way, the self improvement loop, this is the mentality for when an agent will create skills. How cool is this? Now I wanna do a callback to that video I made on Perplexity Computer.
22:56And one of the things I said in that was that because I couldn't change too much with Perplexity Computer, I couldn't tinker with it too much. I was able to just be creative and actually use it instead of sharpening my axe all the time. Hermes, while I can tweak it, it's certainly tweakable.
23:12It's less. It also just kinda works and does its own thing and figures things out, allowing me to just work with it, tell it to do things, and it figures it out. Kinda in that same vein.
23:21And I think that's why it's so successful for most people. Like what Jeff said, we're kinda just getting out of the agent's way. Get get out of the way of, like, them and, like, let the models be
23:32smart, basically. And and just give them the hand like, the the the model is the brain. Right?
23:36We just needed to give it the hands, the feet, the fingers to touch the world in an appropriate way. And then once it has that haptic feedback, really, that's what it is. Like, the harness is the .
23:48It's a weird thought, but it works. Now back to Ron. He found a bunch of stuff.
23:52He wants to do a deeper search. Go for it, dude. Now while Ron's doing that here, I'm gonna go talk to Ron
23:59in, uh, Telegram. Start a new session here, and let's do something fun. Built into Hermes is a home assistant skill.
24:05I use home assistant to run my studio. Let's make some things happen. I'll tell him to enable it first.
24:10I think it's disabled by default. Brilliant. The home assistant wanted to switch on.
24:14I love the way he talks. Let me give him the IP address and key, and let's see what he does. And look at that.
24:19He's he's figuring things out, saving it to his memory. Okay. Let's let's test it.
24:24Got that lamp behind me. Let's tell him to turn it off. Can you turn off Chuck lamp?
24:29See what happens.
24:33He did it. Yes. Let's say turn it on and make it blue.
24:42Also notice I'm talking to Ron in two different places, and he's doing two different things. Ah, he did it. Yes.
24:51Oh, this is fun. One more thing. We have automatic blinds in there.
24:55They're all open. I wanna see if he can do this.
24:59Alright. I'm gonna go film it and see if that happens.
25:03Here we go.
25:10He did it. What happened?
25:16There's one in the kitchen.
25:21He got it. He found it.
25:29I told you IT is magic.
25:33Back on the network side, Ron is wanting some UniFi access. Should we give it to him?
25:38Yes. Let's see what happens. Now notice this is this is what I learned from the, uh, Hermes team, from Jeff.
25:45Rather than the hubris of being like, I know exactly all the skills or I know exactly all the things that you need, Like, the models are smart. Like, like, get out of the way of the models, basically. You know?
25:54Like, they're smart enough if we let them to just be figure out what it is that you want to do. I don't have to bring on this fully onboarded IT admin with skills that I created and everything. No.
26:05I just bring him on like a person and say, hey. You need to manage this thing. Go look at it.
26:10You need to manage this thing. Go look at it. And he'll figure it out.
26:14This is the IT admin I was looking for. Yes.
26:18I'll see. I'm getting you Unifi creds now while you're waiting.
26:25I don't know why I capitalized your, uh, research everything you can about Unifi so you are ready to go. Now notice I'm not pointing these things out because they're not new things to the agent world, but it can delegate tasks.
26:38It can do sub agents. It's a fully functional agent harness. I'm just honing in on the main things that make it awesome.
26:44Also notice you can, um, interrupt steer or queue messages. So if I say something right now by default, it would interrupt him and go, oh, oh, oh, what what does Chuck need?
26:53And instead, I'll just queue and say, here is your UniFi API key. Alright.
27:00He finished his inventory. Realized he couldn't get very far without username and password, but self improvement. Created a UniFi network operation skill.
27:07Let me give him the user account. Now we'll come back to this. We'll see how far he goes.
27:11Well, we gotta move on to number five.
27:16And this one is just funny because it goes back to the idea that OpenClaw was kinda clunky. Jeff's team said it themselves.
27:24This felt a little clunkier compared to what we had internally. And that was my experience. When I initially set up OpenClaw, it was like, oh, this is so cool.
27:32It's so powerful. I don't know about you, but over time, it kind of degraded, and it would break. An update would break it, or it just wouldn't respond.
27:40I'd be like, hey. Are you there? Are you awake?
27:42It became super frustrating, and it wasn't a thing I wanted to give out to my wife or my friends and family. It was one of those things where you had to be an IT person that knew how to troubleshoot it because it was a project. And that's how OpenClaw feels.
27:53It feels like a project, whereas Hermes feels more like a product. And I gotta tell you, I've been using it.
27:59I I I could have made a video a month ago about this, but I wanted to give it some time. I wanted to actually just see if it's cool. I haven't had part I don't think I've had one issue.
28:09Nothing I didn't cause myself. And my wife is having a great time talking to her friend, Honey, who's helping her with homeschool and diet planning and and just managing the house for six kids. Yes.
28:19We have six kids, six daughters. So number five, in case you didn't realize it was that Hermes just doesn't break. The team behind this, they're AI engineers.
28:27And I think that's important and their mentality behind this matters. They're not just gonna add a bunch of features to make it bloated or to compete with somebody else. And sure, it has pretty much feature parity with open claw.
28:38It just doesn't allow a bunch of things that shouldn't be there. They have a philosophy coming into this, which is kinda just get out of the agent's way. Jeff told us earlier, or maybe he didn't.
28:48I don't know we included that quote. We'll get out of the way with the model. Point now to where the AI models we're getting, they're good.
28:54Like, if we stopped getting new models at this point, I think we would be at a point where we could have AGI. What matters the most now is using the right harness, tweaking the tools around it to make it awesome.
29:06Because Opus four dot seven, GPT five dot five, I mean, those things are good. And when you use something like Hermes with it, you really start to realize, oh my gosh. You can do a lot more.
29:15Again, if you wanna hear more about the philosophy behind Hermes, I talked with the cofounder for a while. That interview should be live right now on the academy. Now beyond number five, Hermes is shipping features.
29:26Like, now, there's a Hermes dashboard. Look at this. And if you wanna learn how to do this, by the way, we'll also have that in the, uh, course.
29:32I can only show you so many things here, but check this out. You can get in here and look at your skills and plug ins. You can add more agents, more profiles, add more models.
29:40You can have auxiliary models where you can have one big model that does your main thinking, but maybe have another one that does research or one that's dedicated to delegation. They have achievements, which is kind of fun. Look at this.
29:52Again, the vibe is so awesome. Let them cook. Agent autonomy.
29:57And then one of the most powerful things they released recently is Kanban. You can actually set up a task. Be like, create me as a Pokemon card for network Chuck.
30:04Use my likeness. Make it an HTML page. Make it look awesome.
30:10I'll assign it to default because that's the profile, and we'll create it. And there it goes. How cool is that?
30:17You can jump into it and see the progress. I say hit a rate limit. You can add comments.
30:22It will stop and ask for human input if it needs it, and it got blocked because I hit my usage limit. But you get the picture. Also, just came out their own computer use, which allows it to control your computer.
30:33I think it's still in preview, but, man, they're moving things along. It's becoming very exciting. I'm kind of a fanboy.
30:38I'm being honest. Like, it's it's fun. I like the vibe.
30:41I like the people. I'm not sponsored by them, but I think they're really cool. Let me know what you think.
30:45Like, try Hermes, and you can run OpenClaw and Hermes side by side. You can do your OpenClaw migration. You can do your test.
30:51I convert it. Are you gonna convert? Let me know in the comments below.
30:54That's all I got. I'll catch you guys next time. Hey.
30:58You made it to the end of the video. If you're new here, at the end of my videos, I like to pray for you, my audience. I'm gonna get some more coffee real quick.
31:05Now I know it's a weird thing. A YouTuber praying for his audience. Why?
31:10Well, I believe in the power of prayer even if you don't. And you know what? Just stick around.
31:15You never know what's gonna happen. I'm not trying to force anything on you. I just genuinely care about you, and I believe that this is a powerful thing.
31:21So let's do it. One two three, pray. God, I thank you for the person on the other side of this camera, this screen.
31:30I just thank you for who they are. I thank you that they are passionate about technology, that they're interested in AI, and I pray you bless that. I pray that right now with the way the world is, with AI moving so quickly that you will be able to calm them down, let them not be too stressed, and point them to the good things to use, the things that will elevate them as a human, the things that will encourage them.
31:51And let them not sell their soul to AI. Let them not offload the things that they need to do to AI. Help them figure that out, god.
31:57Equip them with the gifts they need. Bless their families right now. Let their families be, uh, prosperous and blessed and healthy.
32:04Bless their marriages and their children. I pray blessings over them right now. Bless them as they try out these new AI tools, and I pray that through this learning, through this, uh, just tinkering, that this will produce opportunities for them they never even imagined.
32:19Job opportunities, income opportunities, or just breakthroughs in their relationships or just their own thinking.
32:27I pray this over them, God, and I thank you for them. It's in Jesus' name I pray.
32:33Amen. Thank you for letting me do that. I'll catch you guys next time.
The Hook
The bait, then the rug-pull.
NetworkChuck has been running AI agents longer than most, and when he calls something different, his 3M+ subscribers listen. After a month of daily use with Hermes from Nous Research, he is not only switching himself — he handed it to his wife, who named hers Honey. This is the video where he explains why.
Named ideas worth stealing.
20:00concept
The Self-Improvement Loop
Hermes crystallizes successful task executions into reusable skill files. The Curator agent periodically prunes and improves these. The agent gets smarter from your specific workflows, not a generic marketplace.
Steal forAny agent harness design, content about AI systems that learn from usage patterns
12:58concept
Bounded Memory Forces Curation
Hard character limits on agent memory files (1375 and 2200 chars) force the agent to decide what actually matters. Prevents context bloat that degrades agent quality over time.
Steal forCLAUDE.md design, system prompt architecture, AI memory discussions
27:55concept
Product vs Project
A binary for evaluating developer tools: does it feel like something you use (product) or something you maintain (project)?
Steal forEvaluating any dev tool, positioning JoeFlow vs alternatives, any own-your-stack content
23:32metaphor
Harness as Haptic Feedback
Jeff Quesnelle: The harness is the haptic feedback to the model of the world. The LLM is the brain; the agent framework gives it hands and feet to touch reality and receive feedback.
Steal forAI agent explainer content, positioning agent frameworks for non-technical audiences
How they asked for the click.
29:44product
“If you wanna hear more about the philosophy behind Hermes, I talked with the cofounder for a while. That interview should be live right now on the academy.”
Academy pitch earns its place — the interview is genuinely compelling content. Also has YouTube subscribe CTA embedded inside a coffee-break moment at 04:32.
Visual structure at a glance.
