Every Hermes Concept explained for Normal People — Transcript
Full transcript
- 0:00Hermes is the world's most powerful AI
- 0:02agent, but if you're new, it can be
- 0:04incredibly overwhelming. You're probably
- 0:07looking at the app and asking, "How does
- 0:09this make my life easier? What is a
- 0:11habit? What is an operating system? How
- 0:13does this work?" And without knowing the
- 0:15basics, you will never unlock its full
- 0:17capabilities. So, in this video, I'm
- 0:19going to give you the exact tutorial I
- 0:22wish I had when I first started. We'll
- 0:24break down 21 Hermes concepts, starting
- 0:27from extremely simple building to its
- 0:30most powerful features. And by the end
- 0:32of the video, you'll understand this
- 0:34incredibly powerful technology, even if
- 0:37you're a complete beginner. And if
- 0:38you're new, I'm Jack. I built and sold
- 0:40my last tech startup with a gazillion
- 0:41customers. Now, I'm building my own AI
- 0:43startups, and I share here the stuff
- 0:45that actually works. So, if you haven't
- 0:47already, grab that beautiful coffee, and
- 0:50let's dive straight in. So, let's begin.
- 0:51Every Hermes concept explained for
- 0:54normal people. Now, this is going to be
- 0:55the Hermes tutorial. If you just watch
- 0:57this video, you will be able to walk
- 0:59into any room, and people will think,
- 1:01"Dude, who the hell is this guy? Who is
- 1:03this Hermes genius? Does he work for
- 1:05Hermes?" That's what you They're going
- 1:06to feel at the end of the video. Now,
- 1:09Hermes is massive. 271 billion tokens
- 1:14have used, which is around about 240
- 1:16billion, 230 billion words. It's across
- 1:1822 messaging platforms. It is huge.
- 1:21First concept is the foundation, which
- 1:23is basically it is what we call an
- 1:25agent, not a chatbot. This is
- 1:27fundamental to how you understand and
- 1:30think about the Hermes agent. A chatbot
- 1:32tells you how to book a flight. An agent
- 1:36will actually go ahead and book the
- 1:37flight and find it for you. So, think
- 1:39about this. With a chatbot like ChatGPT
- 1:41and Claude, you give it a goal, and it
- 1:43creates a plan. But with an agent, it
- 1:45has tools, and it can actually go ahead
- 1:47and do actions for you. So, for example,
- 1:50what a lot of people don't really
- 1:51understand here, if you think about it
- 1:52like this, right? It's like a smart
- 1:53friend versus is personal assistant.
- 1:56Hermes is a personal assistant. Let me
- 1:59show you what I mean. So, here for
- 2:00example, I have the Hermes agent
- 2:01connected. If you haven't set it up and
- 2:03you want to learn that, I'll put a link
- 2:04on screen to check out. I can say, "Hey
- 2:06there, my man. I'd like to find for me
- 2:08the cheapest flight from Dubai to
- 2:10Toronto. Flying out, let's say, at any
- 2:13point within the next 2 weeks. Let me
- 2:14know what that looks like." You can ask
- 2:16it questions and this will go ahead and
- 2:18use its tools, its agentic capabilities
- 2:21to take actions for me, find
- 2:23information, and enables me to do
- 2:25anything that I like with it. And so,
- 2:26it's done a manual check. I then said,
- 2:28"Find me one flight and open it up for
- 2:29me in the HTML." It looks gorgeous. It
- 2:31went ahead and did that and as you can
- 2:32see, I now have it there. So,
- 2:34conceptually, Hermes is an AI with
- 2:37access to tools. Could be your Gmail,
- 2:39your calendar, many different things
- 2:41that makes it so powerful. So, wherever
- 2:43you are, wherever your phone lives, you
- 2:45can enact things, you can affect change
- 2:47with your AI equipped with these tools.
- 2:49The second concept of Hermes is what it
- 2:51actually is and when you would use it
- 2:53versus using Cloud Code, ChatGPT, Codex,
- 2:57Grok, all these different things. I want
- 2:59you to think of them in this as a
- 3:00framework to understand where Hermes
- 3:02fits in with this. So, think about this.
- 3:03Hermes is like your beautiful Labrador.
- 3:05It's your dog. Cloud Code is your
- 3:06contractor. Um OpenFlow itself is your
- 3:09roommate. And then, Antigravity can be
- 3:11like your your IDE. It's your it's your
- 3:13buddy, the guy that you work with. The
- 3:14best way to think about this is if
- 3:16Hermes is your dog, it lives with you.
- 3:18It knows you. It gets smarter year over
- 3:21year. And Cloud Code is a contractor.
- 3:24It's brilliant for a specific job, but
- 3:26it doesn't necessarily remember you. The
- 3:28core thing about Hermes and the whole
- 3:30point of it is that the more that you
- 3:31use it, the more it gets to know you,
- 3:33the better it gets, and the more
- 3:35specific it is with you. And what it
- 3:37basically means is you can use it
- 3:39wherever you go, wherever your your
- 3:40mobile, you're in a coffee shop, a gym,
- 3:42you can whip out your phone, chat to an
- 3:44agent whose memory is always always
- 3:46improving about you. Now, can I have,
- 3:48for example, Cloud Code learn stuff
- 3:50about me, too? Of course, you can. We
- 3:52can get those memory systems. But I did
- 3:54with Hermes is we can take it anywhere
- 3:55where there's an it enables us to be
- 3:57fully mobile, fully custom, and just
- 3:59unlock some incredible capabilities. So,
- 4:02if I'm at the desktop, I might be using
- 4:04Claude, but if I'm out and about, I'm
- 4:05going to be using Hermes because the
- 4:07more you use it, the better it
- 4:08understands you. And its whole design
- 4:11ethos it's built to live with you and
- 4:13improve with you over time. Which brings
- 4:14us on to, importantly, the third
- 4:16concept, which is one brand, 22 mouths.
- 4:19What do I mean about that? Hermes itself
- 4:22can be accessed from over 22 different
- 4:25interfaces. We use Telegram, but for
- 4:27example, we could talk to it on Discord.
- 4:30We could talk to it on WhatsApp. We
- 4:31could talk to it on any different
- 4:33interface. You can even talk to it on
- 4:35different operating systems. For
- 4:36example, this is my Hermes operating
- 4:38system that connects to my Claude code
- 4:40operating system. And I can ask it
- 4:41questions. I'm like, "Hey there, man.
- 4:43What day is it today?" And I'm just
- 4:44speaking directly to my Hermes agent. I
- 4:46can see every single chat I've had on
- 4:48the left-hand side. And with Hermes, you
- 4:51can actually access the same brand, the
- 4:53same intelligence from all of these
- 4:55different locations. You aren't just
- 4:57constrained to one single input because
- 5:00where the Hermes is running, whether on
- 5:02your computer or VPS, it's the same
- 5:04intelligence and we can connect to it
- 5:06all these different interfaces. For
- 5:08example, if you come and use the Hermes
- 5:10setup code, for example, you'll be able
- 5:12to see all of the different locations
- 5:14that you can actually use with Hermes.
- 5:15For example, this shows you all the
- 5:16different message platforms Slack,
- 5:18Matrix, and all you can do is scroll
- 5:20through these, select them, and then you
- 5:21can connect and chat to the same AI
- 5:23intelligence from any platform that you
- 5:25like. You aren't locked in to just one
- 5:27software. And even in this one here, for
- 5:29example, if you download this Claude
- 5:30code operating system Hermes operating
- 5:32system, you can chat to it in your
- 5:34browser, which I think is a bit more of
- 5:35a dynamic interface. And so, if you
- 5:37think about it as as an analogy, that
- 5:39imagine that you have an operator
- 5:41sitting in the middle, and you can call
- 5:42into this operator from any different
- 5:44one of these interfaces, and she will
- 5:46route every single one of them to you
- 5:47guessed it, your Hermes agent. Which
- 5:50takes us critically onto the next one,
- 5:52which is going to be constant number
- 5:53four, which is where Hermes lives. Now,
- 5:56you've probably heard of it a few
- 5:57different terms like the terminal and
- 5:59VPS and locally hosted. And this is
- 6:02what's really important to understand.
- 6:03The first one is your own computer,
- 6:06which you can run completely for free.
- 6:08This is the easiest, it is the quickest,
- 6:10and in some respects, it's actually
- 6:12incredibly safe to do it this way rather
- 6:14than on other options that I'll show you
- 6:16in a second. You don't need to go and
- 6:17buy some crazy host somewhere. You can
- 6:19do it completely for free on your
- 6:20computer. To do that, all you literally
- 6:22need to do is come over to the Hermes
- 6:23website or click the link below. You
- 6:25copy this, you open up the terminal, and
- 6:27the terminal is basically the way you
- 6:30talk to your computer. It accepts code.
- 6:33And to pull it up, you do command and
- 6:34space bar, you type in terminal, and the
- 6:36terminal will appear. And all you do is
- 6:38control V that command, and effectively,
- 6:41that will install Hermes into your
- 6:42computer. Alternatively, you can go to
- 6:45an app like Cloud Code, and you can say,
- 6:47"Hey there, I'd like you to install
- 6:48Hermes, please. This is the install
- 6:50code. Go and grab it from the GitHub."
- 6:52Or whatever you want to, right? And you
- 6:54can paste it in like that. And
- 6:55literally, Cloud Code will install it
- 6:57for you. So, this is what we mean when
- 6:59we talk about the terminal. Now, it
- 7:01running on your computer just means that
- 7:03as long as your computer is on, Hermes
- 7:05will run. And this is why lots of people
- 7:07have got either like, you know, old
- 7:09MacBooks that they run it on or or
- 7:11MacBook desktop or different kind of
- 7:13laptop, and it just runs for you 24/7 on
- 7:17that computer, or it can just exist on
- 7:19your computer. But effectively, if the
- 7:21laptop shuts down, Hermes doesn't work.
- 7:24So, that's effectively this local
- 7:25hosting revolution. I'm a big believer
- 7:28in local hosting. It's here. It's in my
- 7:30room. It's It's here. I can I can touch
- 7:32it. I can feel it. The second one is a
- 7:34VPS. So, this is essentially where
- 7:36instead of running on your computer,
- 7:38it's actually running on someone else's
- 7:40computer somewhere else in the world.
- 7:42And these companies called hosting
- 7:44providers effectively say, "Hey, what
- 7:46we're going to do is buy like thousands
- 7:48of these like, you know, computers like
- 7:50this, and we're going to stack them up
- 7:51or whatever they're doing over there,
- 7:53and you're basically renting a virtual
- 7:56private server." Okay, so it's running
- 7:58over there. But one of the risks with
- 8:00this is the fact that it it's it's
- 8:01running somewhere else. They can be
- 8:03attacked. You need to do certain things
- 8:04to certify and protect the ports, and
- 8:07you pay monthly for renting that
- 8:10service. For example, our speech attacks
- 8:12at Glyder, that lets me say things and
- 8:14speak as if for instance, you know, we
- 8:16ran some space and for inference to make
- 8:18that fast. So, VPS is an interesting
- 8:20one, and you'll pay monthly for that.
- 8:21And that's what they mean when they say
- 8:22virtual private server. It's running
- 8:25remotely somewhere else. And so, just be
- 8:26aware that these virtual private server
- 8:28companies do pay people money if you
- 8:31sign up using links. So, it's not always
- 8:33necessarily the best thing. Now, this is
- 8:35run here on your computer and free. So,
- 8:37for example, if I come over here, I can
- 8:38say, "Hey there, I'd like to shut down
- 8:41my Hermes agent, please." And you can
- 8:43literally chat to Claude to basically do
- 8:45different things. Now, you'll see as it
- 8:47does that, you'll see my Hermes button
- 8:48now will come down and say, "Gateway is
- 8:50shutting down" as this process goes down
- 8:53and closes for us. And as you can see,
- 8:54just like that, Hermes says, "Gateway
- 8:56shutting down. Your current task will be
- 8:57interrupted." And it shut down, and I
- 8:59can say, "Awesome. Go ahead and reboot
- 9:01him up for me, please." And you can do
- 9:02the same thing. And this can happen
- 9:04because it's running on my own actual
- 9:06computer, which is on 24/7. And this
- 9:08leads us on to the next and fifth
- 9:10concept, which is OAuth versus API. What
- 9:14the hell does that mean? So, there's two
- 9:16ways that you can connect to something
- 9:17essentially. One is what we call open
- 9:19authentication. It is dead straight or
- 9:21forward, and all it basically does is
- 9:23opens up a new browser, and you sign in.
- 9:26Let me show you exactly what I mean. So,
- 9:28here we go. We've got, for example, a
- 9:29terminal. So, we can chat to Hermes.
- 9:31Might come down on my dashboard on the
- 9:33left-hand side to have a look at the
- 9:34Hermes setup command. Let me just go
- 9:36ahead and copy this, throw this in the
- 9:38terminal, and you'll see what I'm in.
- 9:39So, as you can see, it's actually popped
- 9:40up and said Grok, Super Grok, and
- 9:42Premium. Now, for example, if you look
- 9:44at models, which come to shortly, or any
- 9:46servers, we could either do an OAuth.
- 9:47So, for example, if I come down here and
- 9:49I press enter, I might come down and
- 9:51reauthenticate. So, let me come down
- 9:53here and press number two, and you'll
- 9:54see what I'm in. I just want to show
- 9:55you, rather than explaining that. Look,
- 9:57this has opened up a new window. This is
- 10:00open authentication. So, instead of me
- 10:02giving it keys, it just says, "Cool,
- 10:03well, you're signed in, right, bro?" So,
- 10:05why don't you just go ahead and allow
- 10:07me, so you can literally come over here,
- 10:09you click on allow, and then this will
- 10:10now be fully connected to Hermes.
- 10:12Connection successful, and it is easy as
- 10:15that. Now, we're fully connected, and it
- 10:16lets you pick the model. How cool is
- 10:18that, right? Now, if we come and do that
- 10:20again, little hack, by the way, if you
- 10:21press up on terminal, it will actually
- 10:23show you the last thing that you
- 10:24messaged it, which is Hermes setup. And
- 10:26let's say we go now and do a different
- 10:27one. Let's say we want to do OpenRouter.
- 10:29I press spacebar to select it. Okay,
- 10:31come down. It's going to say active
- 10:33provider, and it's asking here for an
- 10:35OpenRouter API key. So, what this means
- 10:37is it needs a physical key. So, you come
- 10:39over to openrouter.ai/models,
- 10:42and from here, we're going to grab a
- 10:43key, and then input it basically here.
- 10:46So, those are the two ways you're ever
- 10:47going to connect. One is open
- 10:49authentication, and the second one is
- 10:51API key. Now, OAuth itself, you can use
- 10:53basically Grok, which is incredible, and
- 10:56you can use ChatGPT, but you cannot use
- 10:59Claude. With OAuth and API key, you can
- 11:02access all the different models, which
- 11:03I'll touch on very shortly. Best way to
- 11:05think about it is that an OAuth is like
- 11:07a sign-in. Okay, so one is a button that
- 11:09you can take back, and the other is a
- 11:11string of characters that live on a
- 11:13server somewhere, and anyone that use an
- 11:15API key with, you can just rotate it
- 11:18anytime you want to. And then once you
- 11:19rotate it in that environment, which is
- 11:22what OpenRouter is so good, it will
- 11:23never work again. So, that's like the
- 11:25big difference. And this takes it on
- 11:26very nicely onto phase two, and actually
- 11:29concept six, which is choosing the right
- 11:31model that that powers us. So, if you
- 11:33think of Hermes as a series of systems,
- 11:36okay? And I'll show you this down and
- 11:37show you exactly what I mean. Hermes
- 11:38itself, right, is an agentic framework
- 11:41with tools and capabilities. But,
- 11:43powering Hermes is a brain. So, for
- 11:45example, in Hermes, I can do {forward
- 11:47slash} type in model like so, and it
- 11:49will let me change the model, the brain
- 11:52behind Hermes. So, at the moment, it's
- 11:54using Grok, but I could switch to Open
- 11:56Router, which gives me access to all of
- 11:58the models on the planet. I can get to
- 12:00Anthropic, I can get to OpenAI Codex, I
- 12:02can do any of these things that I want
- 12:04to, GPT 5.5. Now, if you're using
- 12:06Hermes, you must connect your ChatGPT
- 12:09account to it. And the reason for that
- 12:11is you can use your $20 subscription to
- 12:14code. You cannot do that with Claude.
- 12:17You have to pay for API credits with
- 12:18Claude. So, you're going to see that GPT
- 12:205.5 is like the one of the best ways you
- 12:23can use it there a lot using your
- 12:24existing subscription, and it's exactly
- 12:26the same with Grok, which is why it's so
- 12:28freaking cool. Now, the idea here is
- 12:30that we want to pick the right tool for
- 12:32the right job. I am model agnostic. I
- 12:34believe in the multi-brain strategy,
- 12:37which is basically model agnosticism, or
- 12:39we're agnostic. And effectively, we use
- 12:41the best model for the specific task.
- 12:43So, the best model at reasoning right
- 12:45now, we believe, is Opus 4.7. So, what I
- 12:47do when I want to reason is I come over,
- 12:49I change the model like so, okay? And I
- 12:52will just go ahead and chat to this on a
- 12:53daily basis. I click on Open Router, and
- 12:56I pick Opus 4.7. Now, I pay for credits
- 12:59on this, but in Open Router, when you
- 13:00create an API key, you can set I want
- 13:03this to be $10 maximum per month. That's
- 13:06it. And if it hits $10, it just stops
- 13:09working, and you can go ahead and switch
- 13:10the models, okay? Then you've got, for
- 13:12example, other models that can do
- 13:14volume. So, GPT is an amazing
- 13:16generalist. Grok, again, you can use for
- 13:18volume. Grok, when you authenticate with
- 13:20your X account, can also search Twitter,
- 13:22as well. And then you can bring in other
- 13:24models like Deep Seek, and actually run
- 13:27your entire thing completely for free
- 13:30using a series of free models if that's
- 13:32what you wanted to. So, I want you to
- 13:33think about your brain more as a toolbox
- 13:36rather than a one-size-fits-all because
- 13:38to a hammer everything is a nail, but
- 13:40sometimes you want to use different
- 13:41things like Claude for design and
- 13:43ChatGPT for reviewing code and being
- 13:45more general activities. So, we're just
- 13:46going to tag in the models that we want
- 13:48to and to do that we typically want to
- 13:50use something like OpenRouter because
- 13:52that's one connection that gives you
- 13:54hundreds of different models. little
- 13:55fact by the way, if you come over to
- 13:57openrouter.ai/rankings,
- 13:59you can see month on month by the way
- 14:01who where all where all the things are
- 14:02being used. Look at this. Deepseek V4
- 14:04Flash, right? That's roughly The Flash
- 14:06model is not as powerful as the main
- 14:08model, but effectively you can get like
- 14:0995% performance for 1% of the cost.
- 14:12That's kind of how cool these different
- 14:14systems are. And Concept Seven is a data
- 14:16of what we call local hosted. And local
- 14:19hosted just means that not only is
- 14:20Hermes running your laptop, but the
- 14:22actual model itself is running on your
- 14:25own computer. Now, the trade-off for
- 14:26this is that the biggest models are
- 14:28billions and billions of parameters,
- 14:31which means you need data centers that
- 14:32are huge to run them. So, you're going
- 14:35to be limited based on that, but what
- 14:36you can do on your MacBook, for example,
- 14:37is click the Apple icon at the top,
- 14:39click on about this Mac, and you can see
- 14:41some information about it. So, you can
- 14:43basically screenshot that Mac, you can
- 14:44push that image into Claude or Hermes
- 14:46and say, "Hey, for example, if I come
- 14:48down here and do this real quick, I
- 14:49screenshot this, I share it to Hermes, I
- 14:51can paste the image in and be like,
- 14:53"Hey, based on my MacBook
- 14:54specifications, what do you think would
- 14:56be the most powerful local model that I
- 14:58could run?" Okay? And what we can do
- 15:00that is actually download that model
- 15:02using something called Ollama that lets
- 15:04you download stuff. And you can even do
- 15:06Ollama Cloud, which is a free model
- 15:08hosted somewhere, or I can download it
- 15:10specifically like Gemma 4, Qwen 3.6,
- 15:13whatever, and we can access all those
- 15:14things from your website, which means
- 15:16that whether you are a thousand feet on
- 15:17the ground, you are flying in the air,
- 15:20you're in a rocket ship to space, you
- 15:22will be able to use your Hermes agent
- 15:24anywhere with no limitations, 100%
- 15:27private cuz it is it only exists on your
- 15:30physical laptop. That is the That is the
- 15:32private and local aspect to it. And as
- 15:34you can see, it's come down and it's
- 15:35given us some different ideas like Quant
- 15:3732. B's really powerful. That's 32
- 15:39billion parameters. And these models are
- 15:41just getting better and better. And
- 15:43obviously, because it's local and
- 15:45private, you will compromise a little
- 15:47bit on performance and maybe speed
- 15:49because you're not on a huge server. But
- 15:50that's effectively what people mean when
- 15:52they say run local and run private. Now,
- 15:55we've covered this one. It leads very
- 15:56nicely onto concept number eight, which
- 15:59essentially is never starting from zero.
- 16:02And so, I want you to imagine Hermes as
- 16:04a friend that never forgets what you
- 16:06tell her. And it will remember things
- 16:07like your birthday. And it could tell
- 16:08you things that you spoke about on
- 16:10specific days. This is what we mean when
- 16:13we talk about Hermes' memory. And so,
- 16:14for example, I could say something like,
- 16:16"Hey there, man. Give me an example of a
- 16:18question I asked you on May 19th." Okay,
- 16:21so I can ask actually ask it questions
- 16:23about specific days and it can recall
- 16:25that and let me know based on the way
- 16:27that this memory works. And as you can
- 16:28see, the question I asked it was
- 16:29something about a famous thing about
- 16:31being liked and a really cool
- 16:33interesting quote about throwing rocks
- 16:34at enemies. So, it has within it the
- 16:36Hermes agent does a couple of different
- 16:38memory systems. It has a memory.md. And
- 16:40by the way, MD is just a file type. So,
- 16:43you know how it's like .pptx like
- 16:45PowerPoint or .whatever Word is .x. .md
- 16:49is just markdown file. All it literally
- 16:51is is just text with no special
- 16:53characters. Like that's basically it.
- 16:55It's plain, it's readable, and it lives
- 16:57here on your laptop. Then you've got an
- 16:58SQLite full text search of every message
- 17:01and every session. So, this effectively
- 17:03it can recall everything you've spoken
- 17:05about. Don't need to worry too much
- 17:06about the complicated mumbo jumbo, but
- 17:08just know that it's checking everything.
- 17:09And then the idea is that Hermes agent
- 17:11will compound quietly. So, ideally, if
- 17:13you you need to use it about 2 weeks for
- 17:14it to start feeling magical. And of
- 17:16course, you can supercharge this and
- 17:18connect it to other things like Obsidian
- 17:20memory systems. So, for example, in the
- 17:22Hermes agentic operating system I have,
- 17:24I've actually connected it to my
- 17:26Obsidian system here, which means that
- 17:28all of my Obsidian memory, and I'll put
- 17:30a link on screen if you haven't learned
- 17:31about Obsidian or Pancourt systems, you
- 17:33can check that out. And it also is
- 17:35connected to all of my Claude memory.
- 17:36So, anything I ever talked to Claude
- 17:38Code about, or any of the apps that I
- 17:41use on my computer, I can actually
- 17:42access that and it will dream for me
- 17:44overnight back proactively with
- 17:46suggestions based on questions that I've
- 17:48asked it or how I'm using it or what I'm
- 17:50spending my transcriptions, and it pulls
- 17:52together all these different you get
- 17:54information sources together to give you
- 17:56really specific stuff. To take its
- 17:57memory to a completely interstellar
- 18:00level because if your Hermes agent isn't
- 18:02getting information from stuff you're
- 18:03doing on your computer, Claude Code,
- 18:06Codex, how can it possibly give you the
- 18:08correct and accurate advice? Which is
- 18:10why I think the everything system,
- 18:12bringing it all together into one
- 18:13beautiful operating system, is really,
- 18:15really important, absolutely crucial.
- 18:17And this brings on to the beautiful
- 18:19concept number nine, which is what we
- 18:20call the character Bible, also known as
- 18:23the soul.md. So, this is the file that
- 18:25turns your chatbot into your personal
- 18:27assistant. It's somebody who gets your
- 18:29vibe, your communication style. If
- 18:31you're somebody who says, "Hey dude, hey
- 18:33bro." or however you want to talk and
- 18:35communicate, it specifically explains
- 18:37that to the Hermes agent. And what you
- 18:39want to do basically is give it the
- 18:41information. For example, what is the
- 18:43time? What are your values? What is your
- 18:45style? How do you want them to
- 18:46communicate? And what I'm going to do
- 18:47for you down below is put a full soul.md
- 18:50template that I pulled together based on
- 18:51so many hours of using this stuff. It
- 18:53covers stuff like your identity, mission
- 18:55goals, you know, things about your
- 18:56business, how you want it to address.
- 18:58You want to give that to Hermes and
- 19:00think of it as like the character. Like,
- 19:02do you want to speak to like an honest
- 19:03watching type of character or like a
- 19:05Knight Rider type of character? You can
- 19:07give these soul.mds, and this is what
- 19:09makes your Hermes agent different from
- 19:11everybody else's is because it has a
- 19:13unique soul. Unless we specify what that
- 19:15is, how's it going to know what that
- 19:17looks like? So, super important, give it
- 19:19a soul, and make it fantastical. And
- 19:21again, if you're using an operating
- 19:21system, you can actually also see what
- 19:24those souls are in your own database.
- 19:26Now, concept 10 is the idea that Hermes
- 19:27itself comes off the shelf with many
- 19:29different integrations already locked in
- 19:32that. And then, what we would basically
- 19:34going to do here, guys, is just give it
- 19:35the specific API key for that thing.
- 19:38That could be text to speech, you can
- 19:40connect
- 19:43your entire world to Hermes' agent, and
- 19:46think of Hermes as a Swiss Army knife.
- 19:48For example, if I come back over here,
- 19:50and I open up my Hermes agent, I might
- 19:52say something like, "Hey there, dude.
- 19:54What was the title of the last meeting I
- 19:56was in, and what is one action that I
- 19:58said I was going to take away?" So, for
- 20:00example, I've connected it to Granola.
- 20:02Not a sponsor by any or anything like
- 20:03that, but basically, I use this to
- 20:05capture information on my meetings. It
- 20:07sits in there, and basically, now I can
- 20:09chat to this because I've configured it
- 20:11within my Hermes agent, and it will now
- 20:13allow me to ask any questions about
- 20:15anything. I could say, "Hey, dude. What
- 20:16did I talk about 6 weeks ago for this
- 20:18thing to happen?" Now, as you can see,
- 20:20it's going to come down here and ask my
- 20:21permission to run it. So, we're going to
- 20:22say, "Yeah, that's absolutely fine.
- 20:24We're going to allow it for the
- 20:24session." And what's really cool here,
- 20:25guys, is actually come back and said,
- 20:27"Hey, your your connection's expired,"
- 20:29which is fantastic. I'm going to show
- 20:31exactly how you can add this in for any
- 20:32connection within Hermes cuz this is a
- 20:34great teaching moment. Now, check this
- 20:35out. "Hey, dude. I would like you to
- 20:37give me a terminal command so that I can
- 20:40add my Granola API to your environmental
- 20:44variables." Oh my gosh, Jack, what does
- 20:46that mean? Well, effectively, what we
- 20:48never want to do is publish our API keys
- 20:51in the chat with Hermes agent. Because,
- 20:53again, everything is indexed and saved.
- 20:55So, what we do, especially if it's
- 20:56running on our computer, is we save it
- 20:58somewhere on the computer that it can
- 21:00access, but never actually sits in that.
- 21:02Now, check this one out. Let's come down
- 21:04here. All I'm going to do is copy this
- 21:06right now. I'm going to open up the
- 21:07terminal like so. All right, here's the
- 21:09terminal. Let me exit out of this
- 21:10process. Let me create a brand new one.
- 21:11Okay, beautiful. And what I'm going to
- 21:13do is just paste this one down here. And
- 21:15all it's going to ask me to do is change
- 21:16the API key, which is going to be right
- 21:18here. Then I come over to my app. I
- 21:20click on API keys. And then once you've
- 21:21created it, I've got this even
- 21:23simplified version here. And then you
- 21:24just literally replace this information
- 21:26here with your API key. And you're going
- 21:28to copy that, come to the terminal, and
- 21:30paste it in. That's done. I can say,
- 21:31"Awesome. I've updated that. Please go
- 21:33ahead and try it again." And I wanted to
- 21:35include that in just so you get a sense
- 21:36of how you securely and privately update
- 21:39these, again, API keys within your
- 21:41system. Giving Hermes the ability to
- 21:43connect to anything. Anything that has
- 21:45an API key, you can connect to Hermes in
- 21:47exactly the same way. And what's cool
- 21:49about this is Granola also has an OAuth.
- 21:51So, if you just give it the name of the
- 21:53Granola MCP like this, it will actually
- 21:56ask you to authenticate and sign in. And
- 21:57look at this, guys. Hermes Agent would
- 21:59like to access your account. And the
- 22:01Hermes Agent actually opened up this MCP
- 22:04MCP server for me. So, I come down and
- 22:06allow access. And you can see
- 22:07authorization successful. You can now
- 22:09return to the CLI. And all I did, guys,
- 22:11honestly, I just came down. I copied
- 22:12this page. I just sent the link and
- 22:14shared it straight to Hermes. And as you
- 22:15can see, it's come down here and it's
- 22:17confirmed that it's Farm Blast Meeting,
- 22:18which is a community call I had, and
- 22:20exactly one of the questions that we
- 22:22asked. And that's And it's such a
- 22:23wonderful example of how you can connect
- 22:25your Hermes agent to your entire world.
- 22:27So, I could be anywhere and ask it any
- 22:29question I like. My concept 11 is the
- 22:31fact that Hermes lives on your computer.
- 22:33If you've set it up this way. Which
- 22:35means that not only can it do things
- 22:36like create files on your desktop or
- 22:39organize things on your desktop, it can
- 22:41also take actions on your computer. So,
- 22:44for example, it can do what? It can do
- 22:45browser vision, bash, fuse. What the
- 22:47hell does that mean? It has native
- 22:49Chrome DevTools, visual screen reading,
- 22:51real cursor movements, not a screenshot
- 22:53bar, an actual agency. So, if you're
- 22:55running to the gym and you're like,
- 22:57"Damn, I completely forgot this thing."
- 22:58Basically, Hermes Agent can go ahead and
- 23:00do that. Now, to be fair, this isn't a
- 23:02feature I tend to use a lot, but again,
- 23:04if you just think about the fact that
- 23:06the agent lives on a computer, it can
- 23:07take actions and do various different
- 23:09things for you once you enable it with
- 23:11the tools. Concept 12 is around about
- 23:12using MCPs. MCPs stand for model
- 23:16contacts protocol. So, think of it like
- 23:18this, okay? You have an an API is like
- 23:21the wiring, and the MCP is like the
- 23:23instruction manual around that wiring.
- 23:24So, instead of having like a like, you
- 23:26know, a remote, for example, with a lot
- 23:27of exposed wires, MCPs are basically
- 23:30like instruction packages. It's like a
- 23:32universal remote control that lets you
- 23:34connect Hermes to effectively anything.
- 23:37We just used the Granola MCP to connect
- 23:40to basically my meetings. And we can do
- 23:42the same thing, for example, with Zapier
- 23:44MCP. So, Zapier MCP lets you spin up
- 23:46these servers like so, new MCP server,
- 23:49and I can go ahead and I can connect to
- 23:51anything that I wanted, right? If I come
- 23:52down here, for example, let's say that I
- 23:54want to come down and do other, I come
- 23:56over here and I connect to things that,
- 23:58you know, through this you can't connect
- 23:59to other applications. I can add tools.
- 24:01So, for example, I wanted to give it the
- 24:02ability to read emails, right? Well, how
- 24:04would I do that? Well, we can just
- 24:05basically select all the stuff that you
- 24:07want it to do. The one thing I wouldn't
- 24:08give it access to is the ability to send
- 24:10emails just yet. I would just hold off
- 24:12on that and its ability to reply. And
- 24:14when that's done, I can add these 10
- 24:16tools. I can add in things like Google
- 24:18Drive, I can add in things like whatever
- 24:19I want to, Zapier. It's really cool.
- 24:21Then, all I do, I've got skills over
- 24:23here as a second tab. But, what we're
- 24:25really going to want to focus on here is
- 24:26the connection. And basically, what we
- 24:27can do now is generate a token here. And
- 24:30I've got an MCP server to connect to
- 24:32Zapier, which is awesome. Now, what I
- 24:33can do is come back over to Hermes, and
- 24:35I can say, "Hey there, I'd like to
- 24:36connect to the Zapier MCP server. I've
- 24:39got my tokens. Could you give me a
- 24:40command to run through the terminal so I
- 24:42can add these to your environmental
- 24:44variables?" Again, you can paste it, but
- 24:46I I'd often recommend that you do this
- 24:48in the terminal so it doesn't get
- 24:49recorded in your chat logs. And once
- 24:51you've done that, you can say, "Awesome,
- 24:52I've done. Could you tell me the title
- 24:54of the last email I received?" Again,
- 24:56then it will use an MCP to go ahead and
- 24:58do it. Now, why would you create an MCP
- 25:00versus just using an API key? It is way
- 25:03more what we call token efficient
- 25:05because all the checking it needs to do
- 25:07builds up the cost. An MCP basically
- 25:10explains exactly how it works so AIs can
- 25:13basically just look at the MCP manual
- 25:16and they'll know exactly all the stuff
- 25:17they can do, all the buttons they can
- 25:18press in that individual application.
- 25:20So, it knows how to draft emails, create
- 25:23emails. It has full understanding
- 25:25because it standardizes the use of that
- 25:28incredible tool for the agent. And just
- 25:30like that, we've got all the information
- 25:32that we need from this MCP. Concept 13
- 25:34is understanding muscle memory for the
- 25:36agent. The idea here is that the longer
- 25:38we use Hermes agent, the better it gets.
- 25:40What I mean by that is consistently give
- 25:43it feedback as you go so it can iterate
- 25:46and improve as you're doing it. It has a
- 25:47series of Hermes skills it can pick up
- 25:50the more that you use it. It has various
- 25:52other things it can do and you can even
- 25:53if you want to build out different
- 25:55skills quite easily. So, for example,
- 25:57you can go to this website here, which
- 25:58is agentskills.ai. This gives you a good
- 26:00overview of what skills are if you want
- 26:02to go deeper on that. One of the things
- 26:03that I did actually guys is I built out
- 26:05something called the Pantheon, which
- 26:06effectively enables me, say for example,
- 26:08I want to add a persona, right? I can
- 26:10choose let's go with the alchemist for
- 26:11instance. I can explain its job. I can
- 26:14explain its description, give it a
- 26:15system prompt and then actually choose
- 26:17the model that I want to do it with
- 26:19because again, if it's like very
- 26:20intellectualized, we don't want Albert
- 26:22Einstein, you know, washing our car for
- 26:24example. You know, we want him working
- 26:26on the most impactful tasks. And so, you
- 26:28can build out skills like this. Look, I
- 26:30have a deep reasoning task I use for
- 26:31Claude, but you know, for autopilot
- 26:33jobs, I might use a lower cost model.
- 26:36So, you can build out these skills which
- 26:38give it the ability to do something
- 26:39different. So, for example, I could say
- 26:41something like, "Hey there dude, I'd
- 26:42like to do some research for me, some
- 26:44very deep research on the best country
- 26:46in the world to live. Um you know, spin
- 26:47up a sub agent. Why don't you go ahead
- 26:49and use Deep Seek V4 and ChatGPT 2
- 26:53critically analyze it in a nice little
- 26:54loop with Opus 4.7 reviewing everything.
- 26:58Um, just let's go ahead, use a cheap
- 26:59powerful model to go ahead and do loads
- 27:01of deep research, take some time with
- 27:02it, and then come back and share the
- 27:04results based on everything you know
- 27:05about me. Now, what's cool about this is
- 27:07it can delegate to different models to
- 27:09do different things for me, and it can
- 27:11use that as a skill. Now, it will use
- 27:13here for me the deep research skill that
- 27:15I've already created, which is fantastic
- 27:17and it just enables you to get so much
- 27:19more out of Hermes. Now, if anytime you
- 27:21want to stop it from doing a task, you
- 27:22can do {forward slash} stop, and that
- 27:24will actually basically stop Hermes from
- 27:25doing the thing. The concept 14 is about
- 27:27six keys to make it so unbelievably
- 27:30powerful. So, here we've got we've got
- 27:31Q. Now, what's interesting thing about Q
- 27:33here, okay? Now, Q is this, which means
- 27:36that let's say I ask it a question like,
- 27:38"Hey dude, find out for me the best
- 27:40country for octogenarians. Which country
- 27:43world has the highest number of
- 27:44octogenarians?" Okay? Really cool. I can
- 27:46send it off. Then let's say I want to do
- 27:48a follow-up task, I can do {forward
- 27:50slash} Q. And what this will do is send
- 27:53a prompt on the next turn so it doesn't
- 27:54interrupt. I can also do {forward slash}
- 27:57background, okay? And what {forward
- 27:58slash} background lets me do is
- 28:00effectively ask it let it do different
- 28:02tasks in the background. So, for
- 28:03example, I could come down and do
- 28:04background, and then I might come back
- 28:06and say something like, "Hey, find out
- 28:08which country in the world has the
- 28:10healthiest 35-year-olds or
- 28:1232-year-olds?" Okay? Now, this will run
- 28:15in the background. Cuz normally when you
- 28:17use Hermes agent, basically what's
- 28:18happening is it's it's interrupting. So,
- 28:21this enables us in one chat window to
- 28:23have met multiple streams of thoughts
- 28:26running concurrently by using those two
- 28:28tasks. And as you can see, background
- 28:30task has gone ahead and actually
- 28:31physically got started. You got Kanban,
- 28:33which is for a a task board
- 28:34functionality you can have. Reset will
- 28:36clear it all up. Compress will compress
- 28:39the memory. So, important thing to bear
- 28:41in mind on the memory is that there is a
- 28:42certain size of the window. So, if you
- 28:45have, say for example, 10,000 words, and
- 28:47you ask it a question, every time you
- 28:49ask that chat agent a question, it is
- 28:51literally using all 10,000 words plus
- 28:54your question. And the longer it gets,
- 28:56basically, the lower its performance
- 28:58gets. So, what I do is one session, one
- 29:00window done, you want to reset. So,
- 29:02compressing basically means, "Dude,
- 29:03could you just do me a favor and just
- 29:05summarize everything we talked about?"
- 29:07So, it goes from this to like this,
- 29:09basically, and then you can continue the
- 29:11conversation. And as I showed you
- 29:13earlier, we can switch the models super
- 29:14easily. And what we can go down here and
- 29:16do {forward slash} model, and then we
- 29:18can basically Now we're talking to Opus,
- 29:20we're talking to ChatGPT, we can talk to
- 29:22anyone that we physically like. And I
- 29:23put a lot of the interesting ones down
- 29:25here that I thought were really quite
- 29:26interesting that you may find yourself
- 29:27using, you know, basically depending
- 29:29upon what situation you're in the
- 29:30moment. Now, concept 15 is what we call,
- 29:33basically, safety and security for the
- 29:34agent. These are house rules. Now, these
- 29:36agents have access to many different
- 29:38things, and because of that, I think it
- 29:39was Peter Parker told us this, right?
- 29:41With great power comes great
- 29:42responsibility. Might have been Ben's
- 29:44uncle.
- 29:45>> Remember, with great power
- 29:48comes great responsibility.
- 29:50>> But the point here is we basically don't
- 29:52want to give it the keys to the kingdom
- 29:54because that's responsible years, which
- 29:56is why I don't let it have the ability
- 29:58to send the emails to you. A lot of my
- 29:59buddies have done that, they've had
- 30:00their hands burned in the process, and
- 30:02it will get to the point where it will
- 30:03have less errors than humans. But at the
- 30:05moment, we want to follow what's called
- 30:07the principle of least access. And what
- 30:09that effectively means is Herme Herme's
- 30:11and any agent or system or anything
- 30:12you're building ever. I do this with all
- 30:14my clients. You only ever give it the
- 30:16minimal amount of access that it
- 30:18actually needs. The very, very minimal
- 30:21thing. So, for example, there's certain
- 30:22things that we don't want it to go ahead
- 30:23and do. API keys, never, ever, ever post
- 30:27into any chat like a Herme's especially
- 30:29specifically if it's keeping it. It's
- 30:31not that Herme's is bad person, it's
- 30:33just that if you're backing up your
- 30:34Herme's on a daily basis, we just don't
- 30:36want these keys. It's just one less
- 30:38thing for you to think about. So, if you
- 30:40can actually think about using it
- 30:41responsibly like that and follow this
- 30:43principle of this access, you will
- 30:45generally across all of your AI systems
- 30:47be extremely safe. Now, next thing to
- 30:49think about here in concept 16 is called
- 30:51the North Star and this is what's called
- 30:53the goal function. So, think about like
- 30:55this, okay? Rather than going question
- 30:57answer question answer, the idea of the
- 30:59goal functionality within Hermes is that
- 31:01you give it a North Star for a
- 31:02short-term project. So, I could say,
- 31:04"Hey, I want to create a YouTube
- 31:06thumbnail that is pretty freaking epic."
- 31:09So, for example, I I come over to
- 31:10Hermes, right? And I come down here and
- 31:12why don't I just go ahead and clear this
- 31:14entire session. And so, for example, you
- 31:15come to the Hermes agent and you can
- 31:16give it a prompt like this, which is
- 31:18goal. I would like to decide whether I
- 31:19should take an opportunity A or B.
- 31:21Should I do a big consulting role at
- 31:23Anthropic or do I become head of
- 31:24innovation at OpenAI? I would like it to
- 31:26help me out. Help me make the best
- 31:28decision. And look at this. Now, it's
- 31:29come and said, "Look, I'm going to
- 31:30assign a 20 goal budget as well a 20
- 31:33turn budget to this." And effectively,
- 31:35what will happen now is our entire
- 31:36conversation exists to hit and crush
- 31:40that specific goal. So, it will ask me
- 31:41questions about what's about, what's
- 31:43going on, and it will get there on a
- 31:45mid-term basis. Now, what's really cool
- 31:47here is Hermes will not stop until it
- 31:49hits that decision. But, it does run
- 31:51into one limitation and that's the fact
- 31:53that not all goals we do in life can be
- 31:55solved in 20 turns and sometimes we've
- 31:57got to do things. This is why I built
- 31:59this idea of super goals and I'll put a
- 32:01link on screen if you want to look into
- 32:02this. The idea of super goals is that
- 32:04you explain the goal you want to, use
- 32:06the super goal skill, and effectively,
- 32:08what happens here is Hermes comes and
- 32:10asks you loads of questions and the
- 32:12output of that is like say four to 10
- 32:15actions. Some that you do, some that
- 32:17Hermes does. For example, on this one,
- 32:19which is to launch a Hermes course and
- 32:21get 500 sign ups, one is shape the
- 32:23course promise, build an opt-in engine,
- 32:24write course sequence. Then, it's my
- 32:26turn. I have to go ahead and record the
- 32:28video promo and it lets me mark this
- 32:31task as complete when it's done and then
- 32:33my progress bar goes up. But, if I were
- 32:34to remove this and start from scratch,
- 32:36you'll see right now, look, every
- 32:37hearing is a great goal. I can copy this
- 32:39prompt. I can shoot straight back over
- 32:41to Hermes. What I can do is {slash} and
- 32:43I just come down here and click on stop.
- 32:45As if for instance, say gold pause is
- 32:47what we want to do. And the command I
- 32:48believe is goal clear. This will end the
- 32:50goal. And what I can do is come down,
- 32:52give it this information here, and in
- 32:53effect from now what this does will
- 32:55allow me to have a medium-term goal. And
- 32:57I'll say, "Hey Jack, what is the goal
- 32:58that you want to go ahead and set?" And
- 32:59so say, "Hey dude, I would like to
- 33:01launch a paid course and get 1,000 uh
- 33:04email sign-ups to my new Hermes
- 33:06masterclass." Or whatever the thing is
- 33:08that you want to sell or or help people
- 33:10with. And then it starts to ask
- 33:11questions. I'll say, "Hey, in 4 weeks,
- 33:13okay?" And then it asks you questions
- 33:15dynamically. And I tested this so much,
- 33:17which is why I built all these different
- 33:18specific rules into it. And it's all
- 33:20connected to the dashboard. So when this
- 33:21is done, it will literally load up in
- 33:23the dashboard dynamically. And again,
- 33:24ask you different questions to get
- 33:26context. Cuz without context, how can it
- 33:28know that it's helping you out? Then all
- 33:30you ever do is you land on your mission
- 33:31control, you command it in Hermes, or if
- 33:33you wanted to, you can come down and use
- 33:35Claude. And it's the exact same thing,
- 33:37so you can pick up whenever you want to.
- 33:39Honestly, these mid-term goals, these
- 33:40chief working loops, are so freaking
- 33:42powerful. You can plan your life, hit
- 33:44these mid-term goals using Hermes agent.
- 33:46But concept 17 is about using Hermes
- 33:48agent as a team. So the idea is that we
- 33:50can tag in different models to various
- 33:52different things. You can spin up what
- 33:54is called sub agents. So for example, if
- 33:56I come back over now to Hermes agent,
- 33:58what I might do is just say, "Hey, I'm
- 34:00just going to pause that for a second.
- 34:01What I would like to do is go ahead and
- 34:03research for me the best AI companies to
- 34:05work for on the planet. I'd like you to
- 34:07spin up two sub agents. One is going to
- 34:09go ahead and look at the ones
- 34:11in America, and the other one is going
- 34:12to look at alternative jobs maybe
- 34:14internationally, okay?" And what Hermes
- 34:16can do is spin up sub agents. And what
- 34:20sub agents are is think about like this.
- 34:21You're at Hermes, and then underneath
- 34:22Hermes you have two other agents, each
- 34:25with their own fresh contacts. And it
- 34:27delegates these tasks, these two agents
- 34:29go ahead, they do all the work, and they
- 34:31report back. Now, this is happening in
- 34:34parallel, not sequentially, which means
- 34:36that I could have four, five, six
- 34:38different agents all researching things
- 34:40like a team doing work individually and
- 34:42at the end of the day or the end of the
- 34:43turn, everyone hands in their paper,
- 34:45their research, and instead of waiting,
- 34:47you know, six hours, it's done in one
- 34:48hour instead. This is called
- 34:50parallelization, parallel agents, and
- 34:52gets you significantly further faster
- 34:55using Hermes. I use this strategy all
- 34:57the time to get the most out of the
- 34:58agents. For example, here, Hermes
- 35:00co-founder Teknum, 12 parallel Hermes
- 35:03agent enters. I hope I pronounce his
- 35:04name correctly there. Every day to build
- 35:06Hermes itself, monitoring issues, dog
- 35:08food in the Kanban good board, parallel
- 35:10sub-agents are the power user move. So,
- 35:12even the guy that built Hermes agent
- 35:14used parallel agents to do it. So,
- 35:16again, multiple agents working together
- 35:18in tandem to one objective. Or it could
- 35:20be individual objectives, right? One
- 35:21could be doing research, one could be
- 35:22writing, one designer, one scheduler,
- 35:24and they all put these beautiful folders
- 35:26inside your desk. The concept 18 is so
- 35:29powerful. This was called the heartbeat
- 35:32or the cron job. Cron job sounds so
- 35:34freaking complicated. One of the coolest
- 35:35things you can do is this idea that it
- 35:37keeps working whilst you sleep, the 24/7
- 35:39AAA agent, okay? A heartbeat effectively
- 35:42is something that you set up to happen
- 35:44every single day. It's essentially
- 35:45called a zombie as well, a zombie
- 35:47process, if you like. And effectively,
- 35:49what it will do is you give it something
- 35:51to do. So, effectively, a heartbeat is
- 35:53silent until something dies. So, a
- 35:55sub-agent pings every few seconds,
- 35:57zombie detection, fresh agent reclaimed
- 35:59to task. Per task retries with
- 36:01hallucination recovery. What does that
- 36:03mean? Let me show you what a cron is
- 36:05first, and then I think it'll make a
- 36:06little bit more sense. So, a cron agent
- 36:08here, if I come down here and just say,
- 36:09"Hey, dude, I want you to do two things
- 36:11for me. I want you to send me a reminder
- 36:12in 30 seconds that the sky is blue,
- 36:15okay?" This is a scheduled task. This is
- 36:18basically in a certain amount of time,
- 36:20it is going to come and give me an
- 36:21update. Now, for example, I have a daily
- 36:24morning briefing, okay? And I say, "I
- 36:26want you to go through my um, email
- 36:27list, I want you to go through my
- 36:28calendar, I want you to go through
- 36:30everything I'm doing that day, use all
- 36:32the information that you know about me,
- 36:34and give me, you know, a daily morning
- 36:35brief of, say, I don't know, five things
- 36:38that are relevant. And it can build that
- 36:39for you. So, every day you get that on
- 36:42repeat, basically. So, you can schedule
- 36:43these wonderful things inside Hermie's
- 36:45Agent, get to come to you at 8:00 a.m.
- 36:48every single morning. But, the one that
- 36:49you're going to find yourself using most
- 36:50of the time are these scheduled tasks
- 36:52that make it feel like it's alive
- 36:54because proactively coming to you and
- 36:56asking questions. You could say, "Hey,
- 36:58uh, periodic times randomly come and ask
- 37:00me questions to check in on me, do a lot
- 37:02of other things." Then, we go to concept
- 37:03number 19, which is essentially the
- 37:05dirty secret not a lot of people want to
- 37:06talk about, which is, can you believe,
- 37:0873% of every quest is fixed overhead.
- 37:11So, what we're talking about here is
- 37:13budget, okay? Now, what I mean by that
- 37:15is that, like, if you are using API
- 37:17keys, you can very easily burn through a
- 37:20lot of money, which is why we want to be
- 37:21very specific about the specific models
- 37:23we're using. Now, tokens, you know, um,
- 37:25a parking meter takes maybe dollar
- 37:27coins, um,
- 37:28AI models take tokens. Roughly speaking,
- 37:31it a token, I guess a 10 tokens is about
- 37:34seven words. It's about seven seven 70
- 37:3675%. That's effectively what the
- 37:38translation looks like. So, just make
- 37:40sure that you often clear the session
- 37:42and that you're using the right model
- 37:44for the right job when you're using
- 37:45Hermie's to keep your token cost usage
- 37:48all the way down. And there's loads of
- 37:49mini strategies, like, don't overload it
- 37:51with a bazillion skills that you don't
- 37:53need, don't make it system prompts too
- 37:55long, keep it nice and clean, and one
- 37:58conversation, one goal, then it's always
- 38:00best practice to clear, and refresh, and
- 38:02start again. And just point it out, some
- 38:03guy actually spent 4 million tokens in 2
- 38:06hours of light usage, 21,000 tokens,
- 38:09just asking about the weather because he
- 38:11made a mistake. So, it just gives you a
- 38:12good a good idea of how you can reduce
- 38:14those down. Now, the 20th concept here
- 38:17is this idea that it isn't just a
- 38:19chatbot, Hermie's Agent can be part of
- 38:21an operating system. Well, we have
- 38:23various different personas in the
- 38:25Pantheon. We can have a memory vault
- 38:27that extends its memory capabilities to
- 38:30a significant level, basically enabling
- 38:33you to call upon any email or meeting
- 38:35that you've ever had, any knowledge
- 38:37source from any YouTuber you've ever
- 38:38seen,
- 38:39connected to notebook 11, a
- 38:41gargantillion different things. We can
- 38:44have, for example, so many different
- 38:45tools and effectively when we create an
- 38:47operating system, it just gives it a
- 38:49place to manage everything. I call this
- 38:51a kind of everything of AI. So, if you
- 38:52look at this Claude code operating
- 38:54system, we have our AI spendage, how
- 38:56much money we're making from our skills,
- 38:57we can see our usage in real time,
- 39:00dreaming improvements. Like I said, they
- 39:01will come to you proactively, mission
- 39:03control, our skills and memory systems.
- 39:05Now, in the Hermes operating system
- 39:08here, I can see at a glance all of the
- 39:09connections it has available to it. So,
- 39:11I know exactly what it's connected to. I
- 39:13know what model I'm using, I know my
- 39:14memory, I know my usage. I can chat to
- 39:17it directly. I have this ability to
- 39:19manage goals and plans with Hermes
- 39:21agent. So, whatever I'm working on, I
- 39:22can work with Hermes agent in tandem on
- 39:24that. I have this Pantheon, so I can
- 39:26create these beautiful I don't know I
- 39:27personally like to visually see my
- 39:30graphics and what I'm dealing with. So,
- 39:31I can like, well, actually now I really
- 39:32want to kind of upgrade that like I'm
- 39:34missing this or I'm missing that. I can
- 39:35grab different things and build various
- 39:37different things. I just think it's a
- 39:38nicer user interface to be able to go
- 39:40ahead and build things like that. Again,
- 39:42and what we can do here crucially and I
- 39:43highly recommend you do is connect your
- 39:45Hermes to GitHub. Now, GitHub is a
- 39:48basically code for place where we store
- 39:51stuff. Okay? It's like the world's
- 39:52repository. It's basically you can store
- 39:54information and files. And what you can
- 39:56do with Hermes agent, guys, is
- 39:57essentially have your entire Hermes
- 39:59agent backed up on GitHub every single
- 40:02day. Mine does that. And what that means
- 40:04is if you ever change your computer or
- 40:05God forbid you lose it or something, you
- 40:07have that backed up somewhere, right? We
- 40:09can connect to GitHub. We've got all the
- 40:10profile information here. We can connect
- 40:12to our Obsidian brain, which is this
- 40:14beautiful memory system here that
- 40:16basically just lets us see everything
- 40:18we're doing with different memory
- 40:19systems. We've got there's my Claude and
- 40:21it just enables you to get a level of
- 40:22control. And crucially it comes on to
- 40:25the 21st concept which is the idea of
- 40:28one brand. The idea that we are going to
- 40:30connect the Hermes agent, the business
- 40:33brand, with Claude code. We talked about
- 40:35the differences but here's the
- 40:36difference guys. If we can actually
- 40:37connect them together, it unlocks
- 40:39incredible capabilities. Hermes knows
- 40:41your projects, your customers, your
- 40:43decisions. It lives everywhere.
- 40:45everywhere. Maybe it's like cotton who
- 40:46knows what it is. And then you got
- 40:47Claude code is this precision tool that
- 40:48we use for building. And actually what
- 40:50happens is when we can physically
- 40:52connect these ones together, we have
- 40:54this shared memory and understanding
- 40:56where we're not operating in isolation.
- 40:59It's like you say one thing to Hermes
- 41:01agent, another to Claude code. How do
- 41:03you know they connect together in a
- 41:04meaningful way unless they have access
- 41:07to basically what they're doing at
- 41:08different times. And so now we've
- 41:09covered these 21 core concepts in plain
- 41:13English. The next thing we need to do is
- 41:14leverage all of these and get one of
- 41:16these operating systems ourselves which
- 41:18you can learn about in this video right
- 41:20here.
About this transcript
This page contains the full transcript of Every Hermes Concept explained for Normal People by Jack Roberts, generated from the public captions YouTube serves with the video. The transcript has 9,262 words across 1,291 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.