41:21Jack Roberts · Tutorial
Every Hermes Concept Explained for Normal People
A 41-minute field guide to the open-source AI agent framework — 21 concepts, zero jargon, one tutorial that starts from zero.
May 29thThe argument in one line.
The gap between a chatbot user and an agentic power user is not intelligence — it is configuration depth, and each level of Hermes unlocks a qualitatively different kind of output with compounding time returns.
Read if. Skip if.
READ IF YOU ARE…
- You already use Claude or ChatGPT conversationally but feel like you are doing most of the work yourself.
- You are a solo builder or operator who wants to delegate research, admin, scheduling, and software builds to an agent.
- You are comfortable installing desktop software and connecting API keys.
- You want a model-agnostic setup that can route tasks to Claude, GPT-5, Grok, and DeepSeek based on cost and capability.
SKIP IF…
- You want a quick no-setup automation — Hermes requires intentional configuration at every level.
- You are not interested in running an agent locally on your machine.
- You are a team with existing enterprise tooling that already covers orchestration.
The full version, fast.
Hermes is a local desktop AI agent that takes actions — browsing, writing code, drafting emails, scheduling tasks — while the user is elsewhere. The video walks seven named levels from download-and-go one-shot prompting, through soul.md personalization, model-agnostic routing (Opus for synthesis, DeepSeek for grunt work), MCP integrations (Gmail, Notion, Slack, Granola), parallel sub-agent orchestration, async scheduled builds, and finally a unified agentic OS that shares memory across Hermes, Claude Code, and Obsidian. Most users stall at Level 0 or 1; the real leverage appears at Levels 4-7 when Hermes stops being a chatbot and starts being an operator.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:55
01 · The Seven Levels Explained
Host intro; pyramid graphic introduces all 7 levels; promise that viewers will know exactly where they are and what to fix.
00:55 – 02:48
02 · Level Zero and One: Download and Go
Level 0 = nobody actually uses AI. Level 1 = install Hermes, connect Telegram, run one-shot prompts. Saves ~30 min/day. Biggest mistake: treating it like a vending machine.
02:48 – 04:01
03 · Level Two: The Apprentice
Hermes learns who you are via persistent memory. Can recall conversations from specific days. Saves ~5 hrs/week.
04:01 – 05:27
04 · Building Your Hermes Soul
soul.md profile file: business context, goals, key metrics. Two users asking the same question should get radically different answers.
05:27 – 07:48
05 · Level Three: Commands and Skills
/steer redirects in-progress tasks; /background fires parallel tasks without interrupting. Introduces the Pantheon skill system with named personas.
07:48 – 09:38
06 · Picking the Right Model
Model-agnostic routing: Opus 4.8 for hard reasoning, Sonnet 4.6 for balanced cost, DeepSeek for cheap loops, Grok via OAuth for X/Twitter data, OpenRouter for everything else.
09:38 – 11:24
07 · Skills and Time Savings
Assigning a specific model per skill prevents expensive models from doing grunt work. Demo: competitive outlier research on Instagram saved hours. Level 3 can save up to a day a week.
11:24 – 13:08
08 · Level Four: The Integrator
MCP/API connections unlock email reading, calendar creation, Notion, Slack, ClickUp. Hermes can answer 'what were the numbers this week?' by pulling from connected databases.
13:08 – 14:45
09 · Connecting Granola and MCPs
Live Granola MCP demo: recalls action items from past meetings. Running locally means existing desktop credentials handle authentication automatically.
14:45 – 15:53
10 · The MCP Candy Store Mistake
Too many MCPs balloon the context window. Rule: only connect what you actively use. Email draft-only policy — Hermes drafts but never sends autonomously.
15:53 – 17:01
11 · Level Five: Orchestration
Hermes becomes a manager directing a team of sub-agents. Can run Opus 4.8 as orchestrator with DeepSeek doing the heavy research loop at a fraction of the cost.
17:01 – 18:17
12 · Parallel Sub Agents in Action
Live demo: DeepSeek researcher + GPT-5.5 critic going in a critique loop, Opus 4.8 overseeing. Completely model-agnostic. Key: know your bottleneck before automating.
18:17 – 20:06
13 · Level Six: The Builder
Hermes ships real software artifacts, not just chat responses. Can code from the gym via voice. Anything buildable in Claude Code is buildable in Hermes.
20:06 – 22:17
14 · Scheduled and Async Tasks
Natural-language scheduling: 'remind me midday Saturday Dubai time, do deep research between now and then.' Daily brief runs automatically. Relationship shifts to fully async.
22:17 – 23:33
15 · Level Seven: Agentic Power User
One OS for every AI. Unified dashboard across Hermes, Claude Code, Codex, Antigravity. Hermes proactively analyzes all past interactions across tools to surface gaps.
23:33 – 24:54
16 · One Operating System for Everything
Dashboard shows spend per model, all artifact outputs (PDFs, documents, software), memory graph, Obsidian integration. Memory is no longer siloed per tool.
24:54 – 25:35
17 · Three Truths to Understand
1: Hermes uses Claude Code, it is not competing with it. 2: Cannot realistically run 1,000 agents simultaneously. 3: This is learnable in a couple of videos.
Lines worth screenshotting.
- Hermes is not a competitor to Claude Code — it is a shell that runs Claude Code, GPT, DeepSeek, and Grok inside one agent.
- The soul.md file is what separates a personalized agent from a generic one: two users asking the same question should get radically different answers.
- Connecting too many MCPs bloats the context window — fewer, actively-used integrations outperform a fully-loaded MCP candy store.
- The optimal Level 5 pattern is cheap models (DeepSeek) doing research loops, expensive models (Opus 4.8) doing the final synthesis.
- The async unlock at Level 6 changes the user relationship from call-and-response to things getting built while you sleep.
- Speech dictation is described as a 4-5x speed multiplier for interacting with Hermes — voice is the intended input, not keyboard.
- Scheduled tasks in Hermes use plain English, not cron syntax — precise timezone instructions work as natural language.
- Level 7 is defined by unified memory across tools: Hermes, Claude Code, Obsidian, and Codex share context instead of operating in silos.
- The biggest beginner mistake is automating things that should not exist — automation amplifies a wrong bottleneck if the bottleneck is misidentified.
- MCPs handle authentication via desktop credentials already stored locally, avoiding key duplication across tools.
Seven rules for using an AI agent that actually ships work.
WHAT TO LEARN
Most people stall at Level 1 not because the tool is hard, but because they never configure it to know who they are or what they are trying to do.
- An agent without a personal profile answers generically — a soul.md or equivalent context file is what makes the same question return a different answer for different users.
- Model routing is a cost discipline, not a luxury: assign cheap models to high-volume research loops and expensive models only to final synthesis or hard reasoning.
- Every MCP integration you add expands the context window even when inactive — connect only what you actively use, and treat the MCP list as a budget, not a feature collection.
- The shift from synchronous prompting to asynchronous scheduling is the single biggest unlock: tasks that run overnight or on a defined schedule return results without consuming your attention.
- Parallel sub-agents follow the same division-of-labor logic as a human team: a cheap researcher, a critical reviewer, and an expensive orchestrator who synthesizes the final output.
- Hermes does not compete with Claude Code — it runs Claude Code inside it. Understanding this prevents the most common misconception about what agent wrappers actually are.
- Automating the wrong thing accelerates the wrong outcome: identifying your actual bottleneck before designing an agent workflow is the prerequisite to any useful leverage.
Terms worth knowing.
- Hermes Agent
- A local desktop AI agent that takes real-world actions — file creation, web browsing, code execution, scheduling — via natural language, running models from multiple providers inside a single interface.
- soul.md
- A personal profile file that tells Hermes who you are, your business context, goals, and communication preferences, allowing it to tailor every response to your specific situation rather than answering generically.
- Model agnostic
- The ability to route different tasks to different AI models (Claude, GPT, DeepSeek, Grok) based on cost, capability, or privacy requirements, rather than being locked to one provider.
- MCP (Model Context Protocol)
- A standard for connecting external tools and data sources — email, calendar, Notion, Slack — to an AI agent so it can read and act on them without requiring manual data entry.
- Pantheon
- Hermes's skill system, where you define named AI personas or task specialists each with their own model assignment and system prompt.
- Orchestrator agent
- A top-level agent (typically the most capable model) that manages and routes work to multiple cheaper worker sub-agents running in parallel.
- Asynchronicity
- In the context of Hermes, the shift from synchronous back-and-forth prompting to having the agent execute long-running research and build tasks in the background without waiting for user input.
- Context window bloat
- The degradation in agent performance caused by loading too many MCP integrations or tools simultaneously, consuming tokens on inactive connections and reducing room for useful context.
- Agentic OS
- A unified dashboard where memory, artifacts, spend tracking, and skills from multiple AI tools (Hermes, Claude Code, Codex) are visible and interoperable in one interface.
- OpenRouter
- A service that provides API access to nearly every available AI model through a single key, used here to route tasks to free or cheap models and to access Claude Opus via API.
Things they pointed at.
13:12toolGranola
Lines you could clip.
00:54
“Barely anybody is using AI, let alone Hermes agent — so you are already ahead.”
04:11
“If two people speak to Hermes and ask it the same question, they should get radically different answers because it understands context.”
19:35
“It is no longer just a chatbot. This is software. You are building software.”
24:41
“It is not a better coder than Claude Code. It is literally using Claude Code.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
00:00I have spent hundreds of hours in agent, building, experimenting, and I've seen every single master class. And in this video, I'm gonna take you through the seven levels of Hermes agent step by step. So you know what to expect at each level, the exact mistakes that you need to avoid, and what you need to do to progress to the next level so you can save more time, make more money, and get light years ahead of everybody else.
00:22And if you're new, I'm Jack. I built and saw my latex startup with the gazillion customers. Now I build my own AI companies, and I share here the stuff that actually works.
00:30So if you haven't already, grab that beautiful coffee, and let's dive straight in. Now there are seven levels to this Hermes agent, and if you make it towards the end of the video, they're gonna be saying, hey, who is this super intelligent Hermes individual I see before him? That's what they're gonna be saying about you when you make it to the end of this video.
00:46Now we're gonna go through every level of Hermes from one prompt to an agent that literally does wonderful and fantastical things and a lot of mistakes. Now here's the thing. What level zero is where everybody's at.
00:57Because just for context, nobody's using this stuff. Like, you're seeing all this content because you watch it, but just, like, for context, barely anybody is using AI, let alone Hermes agent, so you are already ahead.
01:08Now let's begin with level one, which is the beginner. This is basically the download and go individual. Essentially, what's happened here is you understood that it's an agent and not a chatbot.
01:19We know that chatbots, you can have conversations with, like Claude and ChatGPT. Hermes is something that actually goes ahead and does things for us. And to reach level one, what you needed to have done is actually gone ahead to basically the website, grab the information, and download Hermes onto your computer.
01:34And if this sounds like I'm speaking German, click this link on screen. I'll show the full process of getting it set up. But effectively, you've gone and you've installed it on your computer.
01:41Now you've installed it on your computer and you're connected to Telegram or your messaging platform of choice. Now at level one, you're using it for simple tasks.
01:49So you might say something like, hey there. I am going from Dubai to a random country in Europe for three months. Do some research.
01:55Write me a list of the best countries to go to with some insights, and just save that HTML file on my desktop so I can check it out for me later when I come back. Send that off, and literally, Hermes agent is gonna do this for us and just save things for us to check whenever we want to.
02:09And as you can see, Hermes agent completes things for us. It uses tools. It's got the chat.
02:14I can just do anything we want to in those current environments, and now I can even open this up and see its output. And as you can see, I open it up and it's giving me a nice little breakdown of all the things that I want to do. Key difference here is that you're using Hermes agent for anything and also tool calling to build interesting things.
02:29Now the biggest drawback at level one is you're just using it for long one shot prompt things. You need to make sure that you're collaborating with Hermes agent.
02:38Ask it to understand your stated intention. Go back and forward and just be mindful of your context window. You could probably expect to save about thirty minutes a day when you're using it in this type of fashion.
02:47And this brings us on to the second level of Hermes agent and one of its strongest features that if you don't use it, you are not utilizing Hermes properly. Now with the apprentice at this level, effectively, Hermes is getting to know you as an individual.
03:02Now the key thing about Hermes agent is that the more that it knows you, the better it actually gets. And I can show you here, for example, this memory system. The more inform the more you speak to Hermes agent, the more information that you give it, the higher performance it will actually get.
03:15But what do I mean by that? Let me explain. So Hermes itself has a very cool memory system.
03:20For example, you can say to Hermes' agent, hey. What was that thing? Could you just tell me one thing that we spoke about two days ago, please?
03:26You can actually get Hermes' agent to recall specifically things that happened on days. Now the cool thing is Hermes will commit to its own internal memory the more that you actually talked about it. But it isn't even just about its memory on a service level because this is just basic at level two.
03:42It's also its ability here to know who you are and tailor its specific advice based on what you're trying to actually get. And as you can see, it's talking about Alama, which is a video that came out a couple of days ago on my channel. Now you are missing out on Hermes magic unless it knows about you.
03:58So what you want to do effectively is use what we call a hermessoul.md. What do I mean by that? Well, if I show you here, for example, in the OS, if I come down here, you can actually categorize your user profile and actually physically give it a salt.
04:10This is stuff like, how do I want you to address me? Information about you and basically your business or whatever it is that you actually are physically doing. And I'll put a link down below for the salt.nd Hermes template that takes through everything, mission goals, your business, key metrics, all this sort of stuff.
04:24The beauty the beautiful part of Hermes, okay, is that if two people speak to Hermes and ask it the same question, they should get radically different answers because it understands context. It's all about context engineering. It needs to know about you and your information, your data so it can best help you.
04:39And that's one of the biggest differentiator is that we start to see at level two on our seven level scale. That's why we call it the apprentice. And the key point here is that Hermes never starts from zero.
04:50It's the agent that remembers your birthday. Its memory is one of its most coolest features, and it should and it should know that stuff about you. So at this level, we should be looking at roughly five hours a week that we're claiming back.
05:00Key thing is if it knows you, you can stop carrying all the information in your head. And remember this, when we're thinking about how much time it saves you, there's such a thing as context switching. Right?
05:10You it isn't just how much time do I spend doing something. You lose things when you switch between tasks. So having one of the world's most intelligent personal AI assistants is an unbelievable game changer and why it's blowing up so much, and it's gonna be such a huge thing in the future.
05:22Key details. It remembers your preferences. It can recall past chats, and there's no need to re explain.
05:27Now custom instructions are one thing, but if you don't use skills and commands properly in level three, you're not utilizing its full potential. What do I physically mean by that?
05:37Well, we're looking at two key things here. Okay? First thing I wanna touch on here is some of the key commands and controls that will radically change your experiences home with Hermes.
05:46At this level, you're starting to basically really utilize its functionality and how cool it freaking is. Let me show you exactly what I mean. First of all, I might give it a prompt that sounds something like this.
05:56Hey there. I would like to go online and find for me what are what is the best advice to scale my LinkedIn and acquire clients in that channel? Do some deep research for me and find that out.
06:05Okay. So you can give Hermes a problem like this. We could be in the gym.
06:09We could be anywhere. Now, crucially, I can do a couple of things. I can do forward slash.
06:13I can do steer. And steer, basically, will guide it in one direction.
06:17I specifically want you to look at top creators on LinkedIn who are already doing very well and also YouTube. Now what this means is I'm kind of like it's it's going to use this and think about this prompt before it's even completed.
06:31Now and it's a great little hack. Another one I can do is background, which basically means whilst it's doing that task, fire this one off in the background.
06:38So I could say, hey there. I'd like to go ahead and do some research on the best way for me to grow my Instagram. Specifically, go and find me great Instagram creators and build me a beautiful list of ways that I can improve my Instagram.
06:49Now background tasks themselves happen in the background, and it doesn't even interact with the core thing that Hermes agent is doing. So the idea here is that we're using more of the features that exist within Hermes. We've got hand off.
07:00We've got clear basically, the functions. And, again, there's lots of different things you can do. Now the other really interesting one here is the model.
07:08Now at this level, at level three, we're basically using what I call model agnostic behavior. What the hell does that mean, Jack?
07:16What it basically means is that we tag in the best model for the job. That's simpler. In fact, I can even show you with the Hermes app.
07:23If I pull this up and even if you see here, for example, in the operating system for Hermes, if I come over here, you can see the model that we currently have tagged in. In my previous video, I also showed you how you can run this 100% privately, which is crazy.
07:36Now the point here is that we can see the model we're dealing with. We're talking to Checkjeet 5.5, and if I have a conversation, I've got that.
07:42Freaking awesome. So for example, if you can come now, you can see the different models I have within Homey's agent. Right?
07:47Anthropic, GitHub, OpenAI Codex, Openreacher, etcetera. So I speak to people all the time about Hermes Agent.
07:53Now some say, Jack, my credit was credits went very quickly. Your core model okay. Think of it like this.
07:58You wanna connect your ChatGPT subscription to Hermes Agent. Super duper important.
08:03That lets you use your $20 ChatGPT subscription for anything. It is a coding powerhouse. Now the most intelligent model is Opus 4.8.
08:13The reality is you don't need it for everything. We don't need astrophysicists, you know, basically building Legos for it, but it's not required.
08:20SONET 4.6 is a very good middle ground of, like, really good performance, but it's not gonna burn your cost. These are API costs. The other model you need to have in here as well is Grok.
08:31Grok also connects via OAuth, which means open authentication. You literally connect it, you sign in, and you're basically using Grok within the engine.
08:40Now the cool thing about Grok, again, it's on your subscription, and you get access to the entirety of x. So you can say, hey.
08:47Uh, what's happening this week with websites? And it'll tell you all the top five posts. It's freaking amazing.
08:52Then we wanna tag in the right models. You wanna make sure that you've got OpenRooter as well. OpenRooter is a very important piece of software that effectively connects to anything in the world that we want to.
09:02Why do I like it? Well, let me show you this. If I come over to rankings, you can see.
09:06We have access to every model. You can access free models. You can access any model that you wish.
09:10It's really freaking impressive. This is how I personally access 4.8, um, with Claude.
09:16And, essentially, we can ratchet it down whenever we want to. We can even use DeepSeek v four Flash, which is fantastic. So we get to basically ratchet up based on performance and cost and also privacy.
09:28So we get to move those dials up to get the right models we want to. And at level three of the system, guys, what we're doing is leveraging the right model for the right task. This is also the level where we begin to introduce skills.
09:41Now when it comes to skills, what I think is really important is the skills are only as good as the model doing the skill and what the skill actually is. So for example, in the Pantheon here, I've got in the operating system. One of the things that I'm very specific about when I add my skills is specifying three things.
09:56So let's go with the alchemist in this case. Right? I specify what the job is, a description of that job, and then a system prompt, which is completely optional.
10:04And then, crucially, I come down and I actually specifically outline what the model is. Why? Because I don't want 4.8, which is really expensive, doing something that we could delegate to a free model or something on DeepSea.
10:15And, of course, we can get this entire thing. Then if I wanna change my prompts, I can visually come in and see them and make adjustments as I wanna as I come and go. Now why is this so important?
10:25This really, if you're down to it at level three, are some of the biggest time saves that you're gonna get because it can do things like go and do my competitive research. Go and find for me all the outliers that exist on Instagram, and you can build software that looks like this. And this, for example, if you're trying to grow on Instagram, as if, for instance, shows you all of the outliers.
10:41I scrubbed this from my community. Why is this so cool? Well, this would be something that you could have spent hours doing.
10:46And this is why this can save you up to a day a week because you can now delegate tasks and automations and research to Hermes, and it will execute for you perfectly. Now if this sounds like I'm actually speaking Portuguese, I'm put a link down below for this full ClawdCode course. It'll take you from foundation and setups, building websites, power features, memory systems, Hermes setup, stuff I have never covered on YouTube before.
11:08I'm I'm gonna put all down below. People mess me up this course all the freaking time. You also get complete immediate access to this entire operating system.
11:14I'll put a link down below so you can check that and upgrade if that's valuable to you. And just FYI, by the way, our price on our community is popping up on the thirteenth. So if you see it ahead of time, we'd love to welcome you.
11:24So we've covered this in level three, and it does lead us very nicely onto the fourth level of Hermes, which is the integrator. Now in the integrator section, this is where really Hermes goes to a completely new level.
11:37And if you look at this graphic as if for instance okay. What we're really doing is connecting it and giving it the ability to connect to your world. Because the more things that you touch and interact with that we can plug into Hermes, the more it can really do.
11:50A lot of the stuff that we've covered so far is you speaking to Hermes, and Hermes doing things like, create me a report that does x or go and research y. In level four, we've really unlocked the new capabilities when we give access to things like reading our emails, things like creating calendar appointments with us and our friends and our our colleagues and potential business partners, looking at Notion for, like, things like our to do lists, plugging into Slack so I can get immediate updates.
12:12For example, maybe you're on ClickUp. Right? One of the things that we could do with ClickUp is have our team, and it's one the things that I implemented, basically giving updates on that.
12:20And now, basically, when I wanna know what's going on in my business, I can say, hey, Hermes. What were the numbers this week? And the really cool thing, and I wanna stress this upon you, okay, is that you give it text, but also numbers.
12:31So if I'm saying, hey, Hermes. I'm thinking about this from a strategic point of view. What are your thoughts on this strategy?
12:38And one of the things my Hermes agent will know to do, depending upon what I'm talking about, is it can go to different databases. Say you're connected on ClickUp or whatever you are. It can go through the latest updates that you had that week from your team.
12:48And so, well, actually, Jack, did you know that, um, you know, Kevin or, I don't know, Sarah m or whatever actually submitted a report that unlocked this new thing that's really relevant for you to know. So the idea is you don't have to rely on your memory. You shouldn't have to rely on your memory.
13:01Plugging Hermes into your world will create new optionality for you. And this is a key thing that everybody's missing. How do I actually plug in new things?
13:09Well, it's so unbelievably straightforward. Let me show you what I mean. So say we have our Hermes agent here.
13:13Right? And, it's even working now in the background answering our things before. I'm gonna come down here.
13:17I'm gonna do a forward slash, and we're gonna stop, which is gonna kill all the background processes, and that'll be the end of it. But let's say I wanna connect them.
13:24So one of the big ones I love to connect to is Granolah. They're not sponsoring this video or anything, but the point with Granolah is they will let you chat to any of your meeting histories. For example, I could say, hey, dude.
13:34Uh, what was the title of the last meeting I had, and can you tell me one action I took from it, please? Now when you connect anything to Hermes agent, sometimes you'll just store the API key on your desktop. Other times, you can literally log in via an MCP, which is freaking awesome.
13:48And bear in mind, guys, this is one of the cool benefits of having this on your desktop. Like, it's already got all of your sign ins and credentials, so you don't need to duplicate this. I mean, I'm not getting paid tens of thousands dollars to sell you some kind of VPS.
14:00I tell you exactly what I do in these situations, which is run it locally on my computer. I don't really see the point or value in running it on a VPS, be honest with you. So it's just really cool and just makes it a lot easier to do things like this.
14:12So we would just come down and we click on allow access in this case. And as you can see down here, one action you took away from AI Automation Coffee was to collaborate with someone in workshop regarding iOS knowledge graph. There you go.
14:21You can see exactly how that's connected, and we even set the one up together. And the other thing to bear in mind is if you're ever not sure how to connect something, just say, hey, Hermes. I wanna give you the ability to do this thing.
14:30What do you need from me to do it? And just follow through the instructions. One little side hack for you, don't ever paste those API keys directly in Hermes.
14:38You can save them on your desktop or it'll give you a terminal command. You can enter that in, and it will throw basically explain everything to you step by step. Now where most people get stuck at this level is they basically give it the MCP candy store.
14:51We give you imagine giving the five year old all the candy in the world in a candy shop, it would go freaking and ballistic. It's the same thing here. Now the more MCPs that we give to Hermes' agent, the bigger its context window actually grows.
15:02Now there are cool tools we can do, like laser loading and stuff, but fundamentally, we don't want to give it loads of garbage that we're not physically using so it's context window doesn't balloon significantly. That's just one to watch out.
15:14An idea is the fewer number of keys we can give it, the better. Now a couple of things I really want to bear in mind with this. First of all, emails, I personally do not give it the ability to send emails.
15:22It can draft them. It can add things to my calendar, but it can't send emails just yet. So I kind of like to keep that one internally rather than sending it out.
15:31At some point, it will be fine. But at the moment, I just like to keep that requirement to myself. And so, really, level four is about crushing your admin.
15:39It should be able to do anything you need it to do. Anything that you would personally do on a laptop, Hermes should be able to do that for you. And if it can't, just ask yourself why it doesn't have that access and just connect it to Hermes.
15:50It's a really good exercise to do and will save you so much time. Now level five is when it starts to get very interesting. We call this level the orchestration level, effectively where Hermes becomes not only an executing agent, but a manager that manages other agents to do different tasks.
16:06It can, in fact, spin up multiple agents, and every single updates that you see with Claude Epis 4.8 can now do this cool thing. It can do ultra work.
16:15Hermes can do the exact same thing. Remember, someone asked me recently, Jack, what's better, Claude or Hermes? It's like asking me what is better, Lewis Hamilton or Ferrari?
16:24They're different. Right? Hermes is the if you think of Hermes as more kind of like the the actual agent system, and then the actual driver, the race car drivers are things like Claude and Cachi PT.
16:34This is more of a system that surrounds it. Now the key thing here is that Hermes itself can use OPUS 4.8 and all of these individual models to do many different things at the exact same time. And we just mentioned, didn't we, about the Pantheon, this idea that we can actually go ahead, if I bring up the right actual tab here, and show you this, all these different systems, these different skills that we wanna build out to do various different things.
16:54But what is really cool about Homey's agent is that we can actually ask it to do multiple things at the same time. Let me show you what I mean. And this is really where the sub agents come to play.
17:03So I can say, hey there. I would like you to create for me a team of agents. I would like one of the sub agents to be call it chat g t 5.5.
17:12I want to use DeepSeek v four, and I'd like to task DeepSeek v four with going and doing some comprehensive research on the best industries to sell websites to. I want g p t four excuse me.
17:25G p t five point five to critique him, and I want them to go around in a loop until they've solved the issue. And then I would like to have Opus four point eight as an orchestrated agent that will oversee it, like his team, spin up as many sub agents as you see fit.
17:37But a lot of the work I want done by the very cheap DeepSeek v four. And as you can see, it's actually going ahead and doing that for us. So we have multiple agents in Hermes working in parallel.
17:47Now this is the reason why this is even better than in some respects than OPUS 4.8 is mainly because we're not just using Claude here. Now, obviously, you can connect these things to Claude, which is fine.
17:57But, again, I get to bring in any model I want to to do anything that I want. You can do this with Inclaud, but I also think it's one of the coolest things that we can do in Hermes, is completely model agnostic. We bring in the best model for the job, and we have these fantastic parallel agents working for us.
18:13Now on level five itself, this is where you can really explode your output if you know what you're applying for. Remember, the biggest hallmark, basically, of a of a beginner automaker is that you automate stuff that simply should not exist.
18:27The idea is this is high leverage when you know the high leverage thing that you're focusing in on. Right?
18:33We don't just talk about AI here. We talk about it in the context of winning, like, life, like, in business, in our physical health, in our relationships.
18:40We wanna be the best version of ourselves, And that means we know the problem that we're solving. AI is an accelerant to that end. So the point is you can start to tank your output if you know what you're focusing on and what your bottleneck is in your particular business or situation.
18:54So it's a real explosive feature. But it brings us very nicely onto the sixth level where it takes one step further, and this one is known as the build up.
19:03So first thing to understand about Hermes Agent is that it does ship real things. Anything. I'm getting fired up right now.
19:09Anything you can build in Cloud Code, you can build in Hermes agent. And honestly, for me, Guy, this was the coolest unlock. To be at the gym, doing my bench press, I used to decline, incline, whatever you wanted to do, and being able to actually code things and build things is a freaking unbelievable.
19:26If you're building software, the ability to have those conversations, spin up the different sub agents, and build software whilst you're doing anything else you want to is literally game changing.
19:36It is no longer just a chatbot. This is software. You're building software.
19:40And I'm telling you right now, that is the future. Building these microsoftwares, these micro apps is insane.
19:45If you have the idea, if you can imagine it, you can build it. That's kind of the point here.
19:50Hermes, let's just do this remotely. And in my view, no other app, like in terms of, like, you know, Codex and Claude, really compete with Hermes' remote ability. You can hack it and make it hacky, but then what you start to realize is you're basically just rebuilding Hermes anyway.
20:04So this is why this is so freaking sick. And at level six as well, you start to use its automation features.
20:09And what do I mean by that? I mean about asking it to do things at certain times. Taking what we call asynchronicity, which means that most people do this.
20:17I answer it, you respond. You you ask a question, I respond. What people in level six do, the builder level, is they're actually adding scheduled tasks.
20:25They're saying, hey, Hermes. Um, I need to go and have a conversation this weekend with, you know, I don't know, McKinsey and Co. Um, do some deep research over the next twenty four hours.
20:33And then on Friday afternoon, make sure you give me a fresh report on my desk, find the email addresses of all the board the members of the board, and actually draft me an email. Or you'll say, and also send me a reminder on Thursday evening that I need to go ahead and do this thing. And you're basically creating your scheduled tasks, and it's unbelievably easy to do.
20:50For example, you just say, hey there. I've got a big meeting on Sunday. Could you just set me a reminder, I don't know, midday at 12:00 on Saturday, Dubai time, to do some preparation for my meeting with NVIDIA?
21:02I want you do some deep research between now and then, find out who the members of the board are, and then also go ahead and find me five interesting things to talk about in that meeting. Whatever the thing is, I can send that off. And I'm just explaining.
21:13I'm yapping in English. By the way, when I'm not using Glider, which I'm using for my speech dictation fast in the world right now, I typically do my voice message when I'm talking to Hermes. And I just yap and I talk because the speed at which you can get information out is four or five times faster, which is why I speech type.
21:28It's, like, honestly, the biggest productivity hack I've had and something I use with Hermes all the time. Now Hermes is setting a reminder to remind me of something. I mean, this sends me a daily brief every single day.
21:37It's freaking incredible. And then it's also gonna do some research and come back and make that reality. Look at this.
21:41Right now, set a one off reminder research for June 6, you know, midday. It's like an executive assistant, but it never forgets. And and it's like the most intelligent thing in the world.
21:50You're only really limited by how you're physically applying Hermes to a situation. And what level six is going to give you is something really powerful, and it's this idea I mentioned of asynchronicity, that it ships without you, it does things without you, and your relationship goes from being, I say, you respond to something completely asynchronous, and it does that.
22:09So things get built even when you're not thinking about them. That's like the huge unlock for level six with Hermes. And that then takes us to the doorstep of the most powerful level of Hermes agent, and that is the agentic power user.
22:24And the idea behind the AgenTic power users, we have one operating system for every AI. What do I physically mean when I say that? Well, I'm best showing you.
22:33So this, for example, is the AgenTic operating system that I use. Now here, obviously, we've got a full overview of all the brands that we're connected to, all the different integrations that we have. We can talk to Hermes.
22:43I've been over the incredible functionalities this has, this ability to create midterm goals for your business. And it breaks down in steps and give you, like, eight different steps of what Hermes is gonna do, what you're gonna do, and vice versa. The Pantheon look at our different skills.
22:58And this is where it gets really interesting. Right? After we've got the memory, we've got the Obsidian integration, we've got all our documents, which is crazy.
23:04If I wanna look at, for example, something Hermes made for me, I can look at this. There is nothing else to do this right now. Like, you're gonna build PDFs.
23:10You're gonna build documents. You're gonna build softwares, but they just live in the chat or they're just filed somewhere.
23:16The ability to actually physically see in a dashboard all the stuff that Hermes does for you is incredible. Like, let's say that you wanna write an invoice to the UFC for your UFC debut. You wanna do a head kick or something like that.
23:26Well, okay. It's great that it's in the chat and it can send things, but wouldn't it I wanna come back and look at it and do various different things. I can do that.
23:33But the real unlock is connecting your world together. Remember, context engineering is one of the biggest trends that you're gonna see. It is exploding.
23:40And the idea here, for me to see my spend across every model, Claude code, codex, antigravity, to have the ability to have basically Hermes' agent . What do I mean by that?
23:51I mean, it's gonna look at every message that I've said to it. It's gonna look at all the stuff that I've done, not only in Hermes, but in Claude code, in antigravity, in codex, in anything else that I'm touching.
24:02And we'll use all of that data, all the skills to proactively think about other ways that I can improve this. Like, what does Jack what is Jack not doing that he should be doing? What are the skills that he's not using that he should be using?
24:15To do research and feed that back to you. And, crucially, the ability and then I'll go here in the memory section right now to see everything from my Obsidian memory that connects everything together and literally even the cold code memory and the Hermes memory. The AgenTic operating system users are individuals that don't operate in silos.
24:32They have their entire AI connectivity in one location. And it basically means that I do coding on Claude code.
24:40Cool. I do coding in Hermes. They're no longer separate.
24:42That gap is completely bridged and unlocks capabilities that people don't even realize actually exist. And at this level, not only is it running without you, it reports and does work for you whilst you are actually physically sleeping.
24:55So three truths you really need to understand. It's not a better coder than ClaudeCode. It is literally using ClaudeCode.
25:01That's really important to understand. The number here is that you can't really have over a thousand agents running simultaneously. The agreed number of agents that we think can be done does vary based on different models, but realistically, it's not gonna be a thousand.
25:14And the magic itself isn't something you need to wait seventeen years to do. This is something you can build and learn in a couple of videos if you just apply yourself. Now understanding where you're at in those seven levels and where they fall apart, the next thing we need to do is embrace the most powerful level, and that's when exactly how to utilize this agentic operating system, which we're gonna do together in this video right here.
The Hook
The bait, then the rug-pull.
Hundreds of hours of agent building distilled into seven levels — Jack Roberts maps the exact progression from a first download to an AI operating system that codes, schedules, and ships work while you sleep.
Named ideas worth stealing.
00:00list
The Seven Levels of Hermes Agent
- Level 1: Beginner — download, install, one-shot prompts (~30 min/day saved)
- Level 2: Apprentice — soul.md personalization, persistent memory (~5 hrs/week saved)
- Level 3: Operator — model routing, /steer, /background, Pantheon skills (up to 1 day/week saved)
- Level 4: Integrator — MCP/API connections, email, calendar, Notion, Slack, ClickUp
- Level 5: Orchestrator — parallel sub-agent teams, orchestrator+worker model pattern
- Level 6: Builder — real software shipped via voice, natural-language async scheduling
- Level 7: Agentic Power User — unified OS across Hermes, Claude Code, Codex, Obsidian
A progressive skill ladder for Hermes agent users, where each level adds a qualitatively different capability and a concrete time-savings unlock.
Steal forCourse or community onboarding structure; product feature gating by user tier
07:48model
Model Routing Heuristic
- Expensive (Opus 4.8): hard reasoning, final synthesis, orchestration
- Mid (Sonnet 4.6): general tasks at controlled API cost
- Cheap (DeepSeek v4, free OpenRouter models): research loops, bulk grunt work
Match model capability to task complexity to avoid burning expensive API credits on simple operations.
Steal forAny multi-agent system where cost management matters
14:45concept
MCP Hygiene Rule
Each MCP integration expands the agent context window whether it is used or not. Only connect integrations you actively use. Dead connections are pure cost with no benefit.
Steal forAny agent integration architecture or tool onboarding guide
How they asked for the click.
VERBAL ASK
10:38product
“I put a link down below for this full Claude Code course. You also get complete immediate access to this entire operating system.”
Mid-video pitch embedded in level 3 section. Natural flow — positioned as a resource for viewers who feel lost, not a hard sell. Community price increase mentioned as urgency trigger.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
