Modern Creator
Why Modern Creator?
Jack Roberts · YouTube

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.

Posted
1 weeks ago
Duration
Format
Tutorial
educational
Views
46.1K
1.4K likes
Big Idea

The argument in one line.

Hermes is not a smarter chatbot but a personal AI operating system that compounds its usefulness the longer you use it — and understanding its 21 building blocks is the difference between occasional help and continuous leverage.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…
  • You have heard of Hermes or open-source AI agents but could not figure out where to start or what the terminology means.
  • You are paying for multiple AI subscriptions and want to know whether a self-hosted agent could route the same work more cheaply.
  • You want to move from typing questions into a chat window to having an AI that takes actions, remembers context across sessions, and runs scheduled tasks while you sleep.
  • You already use Claude Code or ChatGPT and want to understand how a persistent mobile-first agent complements those tools rather than replacing them.
SKIP IF…
  • You want an enterprise or team deployment guide — this is entirely solo-builder focused.
  • You are looking for a code-first deep dive into Hermes internals; the video stays deliberately conceptual and demo-based throughout.
TL;DR

The full version, fast.

Hermes is an AI agent that uses tools to take real actions rather than only generate text. It runs on your own machine and learns your context over time through three layered memory systems. The video covers 21 concepts in ascending complexity: choosing which model to wire in, how MCP servers standardize tool connections, how sub-agents parallelize work, and how cron heartbeats keep tasks running while you sleep. The critical operational insight is that 73 percent of every API request is fixed system-prompt overhead, making session hygiene the single biggest cost lever.

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 →
Chapters

Where the time goes.

00:00 – 01:21

01 · Agent Not A Chatbot

The foundational distinction: a chatbot creates a plan, an agent executes it. Live demo shows Hermes booking a flight and rendering results as HTML.

01:21 – 04:14

02 · Hermes vs Other Tools

Four-animal framework: Hermes is your dog (companion), Claude Code is your contractor (project-scoped), OpenClaw is your roommate, Antigravity is your IDE buddy. Use Hermes mobile; Claude Code at desk.

04:14 – 05:48

03 · One Brain, 22 Mouths

Same Hermes agent accessible from Telegram, Discord, WhatsApp, browser OS, or any of 22+ platforms. One intelligence routed through many interfaces.

05:48 – 09:12

04 · Where Hermes Lives

Three deployment options: laptop (free), VPS (~$5/mo), serverless (cents/month). Start local, never pay until you must.

09:12 – 11:33

05 · OAuth vs API Key

OAuth = one-click sign-in, revocable. API key = secret string, keep out of chat logs, rotate when exposed. Claude requires API key only.

11:33 – 14:20

06 · Pick The Right Model

Model-agnostic multi-brain strategy: Claude/Opus for reasoning, GPT for generalist volume, Grok for search and Twitter, DeepSeek for free high-volume tasks. OpenRouter as the single hub.

14:20 – 16:02

07 · Run It Locally Private

Local model via Ollama = 100% offline, private, no costs. Limited by laptop RAM. Paste your Mac specs into Hermes to ask which quantized model you can run.

16:02 – 18:17

08 · Memory System Explained

Three memory layers: memory.md (persistent facts), SQLite full-text search of every session, Obsidian integration. Two weeks of use before it feels magical.

18:17 – 19:34

09 · Build Your soul.md

The character bible that gives the agent a consistent persona, values, tone, and communication style. What makes your Hermes different from everyone else's.

19:34 – 22:33

10 · Connect Your Entire World

~70 built-in integrations. Demo: Granola meeting-notes MCP connected via OAuth and API key to answer questions about past meetings from anywhere.

22:33 – 23:13

11 · Control Your Computer

Hermes can operate a real browser via Chrome DevTools Protocol — real cursor movements, not screenshots. Native computer use.

23:13 – 25:36

12 · MCPs Simply Explained

API = raw wiring, MCP = instruction manual around the wiring. A universal remote that tells the AI every button each app has — token-efficient and reliable.

25:36 – 27:28

13 · Build Hermes Muscle Memory

Skills and Pantheon personas: assign specialist models to specialist tasks. Sub-agents inherit fresh context per task and report back.

27:28 – 31:49

14 · 6 Power Commands

Six slash commands: /q (queue next prompt), /background (run in parallel), /kanban (task board), /reset (clear session), /compress (summarize context), /model (swap brain).

31:49 – 32:49

15 · Safety And Least Access

Principle of least privilege: only grant minimum access the task needs. Never paste API keys into chat. No send-email permission until error rate is trusted.

32:49 – 33:53

16 · Goals vs Super Goals

Goal function: 20-turn North Star — Hermes won't stop until it decides. Super goals: structured multi-step project plans with human and agent tasks, progress bar, dashboard.

33:53 – 35:35

17 · Spin Up Sub Agents

Parallel sub-agents each get fresh context, work simultaneously, report back. Tuesday-morning work done by Tuesday lunch. Hermes co-founder runs 12 parallel agents daily.

35:35 – 38:03

18 · Heartbeat And Cron Jobs

Heartbeat = zombie watchdog that restarts crashed agents. Cron = natural-language scheduled tasks (no syntax required). Daily morning briefing demo.

38:03 – 38:19

19 · Token Cost Secrets

73% of every request is fixed overhead. One user spent 4M tokens asking about the weather by mistake. Use cheap model for volume, expensive for hard calls.

38:19 – 40:31

20 · Build An Operating System

Hermes as the hub: Pantheon personas, Obsidian memory vault, terminal/gateway, tool workshop. GitHub backup daily. The everything system for your AI life.

40:31 – 41:21

21 · Connect Hermes To Claude

Hermes (business brain, long-term memory) + Claude Code (precision builder) share context so neither operates in isolation.

Atomic Insights

Lines worth screenshotting.

  • A chatbot creates a plan; an agent executes it — that one distinction changes everything about how you prompt and what you expect.
  • Hermes is designed to live with you: the more sessions you run through it, the more accurately it models your preferences, habits, and goals.
  • Start local on your laptop for free — never pay for a VPS until you have outgrown your laptop, because most people never do.
  • 73 percent of every AI API request is fixed system-prompt overhead, which means session length is a direct cost multiplier, not just a UX issue.
  • OAuth gives you a revocable button; an API key gives you a secret string — never paste either into a chat that logs its history.
  • Claude requires an API key and charges per token; Grok and ChatGPT can authenticate via OAuth using existing paid subscriptions.
  • The right model for volume is different from the right model for reasoning — treating them as interchangeable is the fastest way to overspend.
  • MCP servers are instruction manuals wrapped around API wiring — they let the agent know every button an app has without burning tokens discovering it.
  • One session, one goal — then compress or reset; the longer a context window grows, the more every new query costs and the lower quality the responses become.
  • Parallel sub-agents each get fresh context and work simultaneously, turning hours of sequential research into a single synchronized handoff.
  • A heartbeat watchdog pings every few seconds and automatically restarts crashed sub-agents so no task is silently dropped.
  • Cron jobs in Hermes require no cron syntax — you write a natural-language schedule and the agent handles the timing itself.
  • The soul.md character bible is what makes your Hermes agent different from everyone else's: without it the agent has no voice, no values, no consistent way of addressing you.
  • Connecting Hermes to Claude Code means neither tool operates in isolation — shared context compounds both.
  • Principle of least access applied to AI agents: only grant the minimum permissions a specific task requires, then revoke when done.
Takeaway

The 21 concepts that turn Hermes from confusing to essential.

WHAT TO LEARN

Hermes is not a better chatbot — it is a personal operating system that gets smarter the longer you use it, and these 21 concepts are the mental models that unlock the difference between occasional help and daily leverage.

01Agent Not A Chatbot
  • An agent differs from a chatbot in one fundamental way: it has tools and takes real actions, not just generates plans — which means the gap between telling you and doing it for you is the whole product.
02Hermes vs Other Tools
  • Hermes is optimized to compound over time with you — mobile, persistent, personal — while Claude Code is optimized for precision on a specific project; using both is not redundant, it is additive.
03One Brain, 22 Mouths
  • The same agent intelligence is accessible from 22 or more platforms simultaneously — your entry point does not determine what the agent knows or can do.
04Where Hermes Lives
  • Start running Hermes on your own laptop for free; only move to a VPS when uptime genuinely matters to your workflow, because most use cases never require it.
05OAuth vs API Key
  • OAuth and API keys are the only two ways you will ever connect a service; treat API keys as secrets that live in environment variables, never in chat logs that get indexed and stored.
06Pick The Right Model
  • Model selection is not brand loyalty — it is task matching: use a reasoning model for hard problems, a cheap volume model for loops, and a free model where accuracy is not critical.
07Run It Locally Private
  • Running a local model via Ollama means zero API cost and full privacy; the tradeoff is raw capability, which shrinks with every billion parameters you cannot fit in your laptop RAM.
08Memory System Explained
  • Hermes memory compounds across three layers — a plain-text facts file, full-text search of every session, and optional Obsidian integration — and takes roughly two weeks of regular use to feel genuinely magical.
09Build Your soul.md
  • The soul.md character bible is what separates your agent from a generic chatbot; without explicit values, communication style, and goals written into it, the agent has no consistent voice.
10Connect Your Entire World
  • Connecting external tools via MCP or API gives the agent access to your real world — meetings, emails, files — which is what makes responses accurate to your actual situation, not generic.
11Control Your Computer
  • Computer use via Chrome DevTools Protocol means the agent moves the real cursor and clicks real buttons — not screenshots — which makes it usable for tasks you cannot stay at your desk to complete.
12MCPs Simply Explained
  • MCP servers are more token-efficient than raw API calls because the agent already knows every button the app has — it does not need to discover capabilities mid-task.
13Build Hermes Muscle Memory
  • Consistent feedback and purpose-built skills let Hermes delegate to the right model for each sub-task — the Pantheon persona system is how you build a specialist team without hiring anyone.
146 Power Commands
  • One session, one goal — then compress or reset; context windows that grow unchecked cost exponentially more and produce progressively worse results.
  • The six slash commands are what separate users who run Hermes manually from users who run it efficiently.
15Safety And Least Access
  • The principle of least access applied to agents means granting only what the current task needs and revoking it after; even the host does not give his agent permission to send emails yet.
16Goals vs Super Goals
  • The goal function turns a freeform chat session into a directed project with a defined end state — super goals extend this to multi-week plans with mixed human and agent tasks tracked on a dashboard.
17Spin Up Sub Agents
  • Parallel sub-agents with fresh context each can cut multi-hour sequential research into a single synchronized handoff — treating the agent as a manager, not a single worker, is the power-user shift.
18Heartbeat And Cron Jobs
  • Cron jobs and heartbeats together give your agent a 24/7 presence: heartbeat restarts crashed agents silently, cron fires scheduled tasks in natural language without any syntax.
19Token Cost Secrets
  • 73 percent of every API request is fixed system-prompt overhead — which means session hygiene is the single biggest cost lever available to you.
20Build An Operating System
  • An AI operating system is not complexity for its own sake; it is the infrastructure that prevents your agent from starting from zero every session.
21Connect Hermes To Claude
  • Connecting Hermes long-term memory and mobile reach to Claude Code precision building means neither tool operates in isolation — shared context makes both more accurate.
Glossary

Terms worth knowing.

Agent
An AI system that uses external tools to take real-world actions such as searching, booking, or writing files, rather than only generating text responses.
Hermes
An open-source AI agent framework that runs on your own machine, connects to 22 or more messaging platforms, and accumulates memory about you over time.
OAuth
Open Authentication — a one-click sign-in flow that grants an app access to your account without exposing a raw key; access can be revoked from a dashboard button.
API key
A secret string of characters that authenticates your requests to an AI provider; must be kept out of chat logs and rotated if exposed.
MCP (Model Context Protocol)
A standardized instruction package that tells an AI agent exactly what actions a given tool or app supports, making tool use faster and cheaper than raw API calls.
OpenRouter
A routing layer that gives a single API connection access to hundreds of AI models from different providers, with per-key monthly spending caps.
VPS (Virtual Private Server)
A rented server that keeps Hermes running around the clock even when your laptop is off; typically costs five to ten dollars per month.
soul.md
A markdown file that defines an AI agent's persona, values, communication style, and goals — the character bible that makes the agent behave consistently as yours.
Sub-agent
A child agent spun up by the main Hermes instance with its own fresh context window; multiple sub-agents run in parallel and report results back to the parent.
Heartbeat
A watchdog process that pings sub-agents every few seconds and automatically reclaims and restarts any that have crashed or gone silent.
Cron job
A scheduled recurring task; in Hermes these are written in plain English rather than traditional cron syntax.
Context window
The total amount of text an AI model can process in one session; as a conversation grows, every new message re-reads the entire history, increasing cost and reducing quality.
Ollama
A tool that downloads and runs open-source AI models locally on your machine, enabling fully private offline inference without API costs.
Pantheon
A Hermes UI feature that lets you create and manage named AI personas, each with its own system prompt and assigned model, for different task types.
Principle of least access
A security practice of granting an AI agent only the minimum permissions a specific task requires, reducing the blast radius of any error or misuse.
Resources

Things they pointed at.

01:21productHermes AI Agent (open source) ↗
11:33toolOpenRouter ↗
14:20toolOllama ↗
19:34productGranola (meeting notes) ↗
23:13toolZapier MCP ↗
16:02toolObsidian ↗
25:36linkagentskills.im ↗
38:19toolGitHub ↗
Quotables

Lines you could clip.

01:28
“A chatbot tells you how to book a flight. An agent will actually go ahead and book the flight and find it for you.”
Definitive one-sentence distinction between chatbot and agent — works standalone with zero setup→ TikTok hook↗ Tweet quote
04:01
“The more you use it, the better it understands you. Its whole design ethos is built to live with you and improve with you over time.”
Captures the compounding-value thesis in two sentences→ IG reel cold open↗ Tweet quote
37:09
“73% of every request is fixed overhead.”
Counterintuitive stat most AI users have never heard→ newsletter pull-quote↗ Tweet quote
37:55
“Some guy actually spent 4,000,000 tokens in two hours of light usage, 21,000 tokens just asking about the weather because he made a mistake.”
Concrete horror story that makes abstract token cost visceral and memorable→ TikTok hook↗ Tweet quote
30:06
“Tuesday morning's work, done by Tuesday lunch.”
Punchy parallelization payoff line — no setup needed→ IG reel cold open↗ Tweet quote
The Script

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.

00:00Hermes is the world's most powerful AI agent. But if you're new, it can be incredibly overwhelming.
00:06You're probably looking at the app and asking, how does this make my life easier? What is a hobby? What is an operating system?
00:13How does this work? And without knowing the basics, you will never unlock its full capabilities. So in this video, I'm gonna give you the exact tutorial I wish I had when I first started.
00:24We'll break down 21 Hermes concepts starting from extremely simple building to its most powerful features.
00:32And by the end of the video, you'll understand this incredibly powerful technology even if you're a complete beginner. And if you knew, I'm Jack. I built and sold my last tech startup with a gazillion customers.
00:42Now I'm building my own AI startups and I share here the stuff that actually works. So if you haven't already, grab that beautiful coffee and let's dive straight in. So let's begin every Hermes concept explained for normal people.
00:55Now this is gonna be the Hermes tutorial. If you just watch this video, you will be able to walk into any room and people will think, dude, who the hell is this guy? Who is this Hermes genius?
01:05Does he work for Hermes? That's what they're gonna feel at the end of the video. Now, Hermes is massive, 271,000,000,000 tokens have used, which is round about 240,000,000,000, 230,000,000,000 words.
01:18It's across 22 messaging platforms. It is huge. First concept is the foundation, which is basically, it is what we call an agent not a chatbot.
01:27This is fundamental to how you understand and think about the Hermes agent. A chatbot tells you how to book a flight.
01:35An agent will actually go ahead and book the flight and find it for you. So think about this, with a chatbot like ChatGPT and Claude, you give it a goal and it creates a plan. But with an agent, has tools and it can actually go ahead and do actions for you.
01:49So for example, what a lot of people don't really understand here, if you think about it like this, right, it's like a smart friend versus a personal assistant. Hermes is a personal assistant. Let me show you what I mean.
02:00So here for example, I have the Hermes agent connected. If you haven't set it up and you wanna learn that, I'll put a link on screen to check out. I can say, hey there, my man.
02:07I'd like to find for me the cheapest flight from Dubai to Toronto. Flying out, let's say, at any point within the next two weeks, let me know what that looks like. You can ask it questions, and this will go ahead and use its tools, its agentic capabilities to take actions for me, find that information, and enables me to do anything that I like with it.
02:26And so it's done a manual check. I then said, find me one fly and open it up for me in a HTML that looks gorgeous. I went ahead and did that, and as you can see, I now have it there.
02:34So conceptually, Hermes is an AI with access to tools, could be your Gmail, your calendar, many different things that makes it so powerful. So wherever you are, wherever your phone lives, you can enact things.
02:46You can effect change with your AI equipped with these tools. The second concept of Hermes is what it actually is and when you would use it versus using Callcode, ChatGPT, Codex, Grok, all these different things.
02:59I want you to think of them in this as a framework to understand why Hermes fits in with us. So think about this. Hermes is like your beautiful Labrador, it's your dog.
03:06Clawd Code is your contractor. Open Clawd itself is your roommate, and then anti gravity can be like your your IDE, it's your it's your buddy, the guy that you work with. The best way to think about this is if Hermes is your dog, it lives with you, it knows you, it gets smarter year over year, and Clerkode is a contractor.
03:24It's brilliant for a specific job, but it doesn't necessarily remember you. The cool thing about Hermes and the whole point of it is that the more that you use it, the more it gets to know you, the better it gets and the more specific it is with you.
03:37And what it basically means is you can use it wherever you go, wherever you're you're mobile, you're in a coffee shop, a gym, you can whip out your phone, chat to an agent whose memory is always always improving about you. Now can I have, for example, Claude Code learn stuff about me too? Of course, you can.
03:52We can get those memory systems. But I did with Hermes is we can take it anywhere with us and it enables us to be fully mobile, fully custom, and just unlock some incredible capabilities.
04:02So if I'm at the desktop, I might be using Claude. But if I'm out and about, I'm gonna be using Hermes because the more you use it, the better it understands you. And its whole design ethos is built to live with you and improve with you over time.
04:14Which brings us on to importantly, the third concept, which is one brain, 22 mouths. What do I mean about that?
04:21The Hermes itself can be accessed from over 22 different interfaces. We use Telegram, but for example, we could talk to it on Discord. We could talk to it on WhatsApp.
04:31We could talk to it on any different interface. You can even talk to it on different operating systems. For example, this is my Hermes operating system that connects to my Cloud Code operating system and I can ask the questions.
04:42I'm like, hey there man, what day is it today? And I'm just speaking directly to my Hermes agent. I can see every single chat I've had on the left hand side.
04:49And with Hermes, you can actually access the same brain, the same intelligence from all these different locations.
04:57You aren't just constrained to one single input because where the Hermes is running, whether on your computer or VPS, it's the same intelligence and we can connect to all these different interfaces.
05:09For example, if you come and use the Hermes setup code for example, you'll be able to see all of the different locations that you can actually use with Hermes. For example, this shows you all the different messaging platform, Slack, Matrix, and all you can do is scroll through these, select them, and then you can connect and chat to the same AI intelligence from any platform that you like.
05:26You aren't locked in to just one software. And even in this one here, for example, if you download this corporate operating system, it's Hermes operating system, you can talk to it in your browser, which I think is a bit more of a dynamic interface. And so if you think about it as as an analogy, that imagine that you have an operator sat in the middle and you can call into this operator from any different one of these interfaces and she will route every single one of them to, you guessed it, your Hermes agent.
05:49Which takes us critically on to the next one, which is going to be concept number four, which is where Hermes lives. Now you've probably heard of a a few different terms like the terminal and VPS and locally hosted. And this is what's really important to understand.
06:04The first one is your own computer, which you can run completely for free. This is the easiest, it is the quickest, and in some respects, it's actually incredibly safe to do it this way rather than on other options I'll show in a second.
06:16You don't need to go and buy some crazy host somewhere. You can do it completely for free on your computer. To do that, all you literally need to do is come over to the Hermes website or click the link below.
06:25You copy this, you open up the terminal, and the terminal is basically the way you talk to your computer. It accepts code. And to pull it up, you do command and space bar, you type in terminal and the terminal will appear.
06:37And all you do is control v that command and effectively that will install Hermes into your computer. Alternatively, you can go over to an app like ClawdCode and you can say, hey there, I'd like you to install Hermes please.
06:49This is the install code. Go and grab it from the gear tub or whatever you want to. Right?
06:53And you can paste it in like that. And literally, ClawdCode will install it for you. So this is what we mean when we talk about the terminal.
07:00Now it running on your computer just means that as long as your computer is on, Hermes will run. And this is why lots of people have got either like, you know, old MacBooks that they'd run it on or MacBook desktop or different kind of laptop and it just runs for you twenty four seven on that computer or it can just exist on your computer.
07:20But effectively, if the laptop shuts down, Hermes doesn't work. So that's effectively this local hosting revolution. I'm a big believer in local hosting.
07:29It's here. It's in my room. It's it's here.
07:31I can I can touch it? I can feel it. The second one is a VPS.
07:35So this is essentially where instead of running on your computer, it's actually running on someone else's computer somewhere else in the world. And these companies called hosting providers effectively say, hey, what we're gonna do is buy like thousands of these like, you know, computers like this and we're gonna stack them up or whatever they're doing over there.
07:53And you're basically renting a virtual private server. Okay.
07:57So it's running over that. But one of the risks with this is the fact that it's it's running somewhere else. They can be attacked.
08:03You need to do certain things to certify and protect the ports and you pay monthly for renting that service.
08:10For example, our speech to text app Glider that lets me say things and speak as if for instance, you know, we rent some space for inference to make that fast. So VPS is an interesting one and you pay monthly for that. And that's what they mean when it's a virtual private server.
08:24It's running remotely somewhere else. And so just be aware that these virtual private server companies do pay people money if you sign up using links. So it's not always necessarily the best thing.
08:34Now, this is run here on your computer free. So for example, if I come over here, I can say, hey there, I'd like to shut down my Hermes agent please. And you can literally chat to Claude to basically do different things.
08:46Then you'll see, as it does there, you'll see my Hermes button now will come down and say gateway is shutting down as this process goes down and closes for us. And as you can see just like that, Hermes says gateway shutting down, your current task will be interrupted and it shut down. And I can say, awesome, go ahead and reboot him up for me please and you can do the same thing.
09:03And this can happen because it's running on my own actual computer which is on twenty four seven. And this leads us on to the next and fifth concept which is OAuth versus API.
09:14What the hell does that mean? So there's two ways that you can connect to something essential. One is what we call open authentication.
09:20It is dead straight or forward And all it basically does is opens up a new browser and you sign in. Let me show you exactly what I'm in. So here we go.
09:28We've got for example a terminal, so we can chat to Hermes. Might come down on my dashboard on the left hand side to have a look at the Hermes setup command. Let me just go ahead and copy those, throw this into terminal, and you'll see what I mean.
09:39So as you can see, it's actually popped up and said Grok, super Grok and premium. Now, for example, if you look at models which come to shortly or any service, we could either do an LWAF. So for example, if I come down here and I press enter, I might come down and re authenticate.
09:52So let me come down here and press number two and you'll see what I mean. I just wanna show you rather than explaining. Now look, this has opened up a new window.
09:59This is open authentication. So instead of me giving it keys, it just says cool. Well, you're signed in.
10:04Right, bro? So why don't you just go ahead and allow me? So you can literally come over here, you click on allow and then this will now be fully connected to Hermes.
10:12Connection successful and it is easy as that. Now we are fully connected and it lets you pick the model. How cool is that?
10:19Right. Now if we come and do that again, little hack by the way, if you press up on terminal, it will actually show you the last thing that you messaged it, which is how we set up. And let's say we go now and do a different one.
10:27Let's say we wanna do open router. I press space bar to select it. Okay.
10:31Come down. It's gonna say active provider, and it's asking here for an OpenRooter API key. So what this means is it needs a physical key.
10:39So you come over to open router.ai/models, and from here, we're gonna grab a key and then input it basically here. So those are the two ways you're ever gonna connect.
10:48One is open authentication and the second one is API key. Now, OAuth itself, you can use basically Grok, which is incredible and you can use ChatGPT, but you cannot use Claude via AWAUF.
11:01An API key, you can access all the different models which I'll touch on very shortly. Best way to think about it is that an AWAUF is like a sign in. Okay.
11:08So one is a button that you can take back and the other is a string of characters that live on a server somewhere. And anyone that use an API key with, you can just rotate it anytime you want to. And then once you rotate it in that environment, is why OpenWood is so good, it will never work again.
11:24So that's like the big difference. And this takes us on very nicely onto phase two and actually concept six, which is choosing the right model that that powers this.
11:33So if you think of Hermes as a series of systems. Okay? And I'll show this down and show exactly what I mean.
11:38Hermes itself, right, is an agentic framework with tools and capabilities. But powering Hermes is a brand. So for example, in Hermes, I can do forward slash, type in model like so, and it will let me change the model, the brain behind Hermes.
11:53So at the moment, it's using Grok, but I could switch to OpenRooter, which gives me access to all the models on the planet. I could get to Anthropic. I could get to OpenAI Codex.
12:02I could do any of these things that I want to. GPT 5.5. Now if you're using Hermes, you must connect your ChatGPT account to it.
12:10And the reason for that is you can use your $20 subscription to code. You cannot do that with Claude.
12:17You have to pay for API credits with Claude. So you're gonna see that GBT 5.5 is like one of the best ways you can use it via OAuth using your existing subscription.
12:26And it's exactly the same with Grok, which is why it's so freaking cool. Now the idea here is that we wanna pick the right tool for the right job. I am model agnostic.
12:34I believe in the multi brain strategy, which is basically model agnostic. And effectively, we use the best model for the specific task. So the best model at reasoning right now, we believe is Opus 4.7.
12:47So what I do when I wanna reason is I come over, I change the model, like so. Okay. And I will just go ahead and chat to this on a daily basis.
12:54I click on OpenRooter and I pick Opus 4.7. Now I pay for credits on this, but in OpenRooter, when you create an API key, you can set, I want this to be $10 maximum per month. That's it.
13:07And if it hits $10, it just stops working and you can go ahead and switch the models. Okay?
13:11Then you've got, for example, other models that can do volume. So GPT is an amazing generalist. Grok, again, you can use for volume.
13:19Grok, when you authenticate with your x account, can also search Twitter as well. And then you can bring in other models like DeepSeek and actually run your entire thing completely for free using a series of free models if that's what you wanted to.
13:33So I want you to think about your brain more as a toolbox rather than a one size fits all because to a hammer, everything is an out. But sometimes we wanna use different things like code for design and chattypity for reviewing code and be more general activities.
13:46So we're just gonna tag in the models that we want to. And to do that, we typically wanna use something like OpenRooter because that's one connection that gives you hundreds of different models. And cool little fact, by the way, you come over to openroot.ai/rankings, you can see month on month, by the way, who where all where all the things are being used.
14:03Look at this. DeepSeek v four Flash. Right?
14:05That's roughly the Flash portal is not as powerful as the main model, but effectively, you can get like 95% performance for 1% of the cost. That's kind of how cool these different systems are.
14:15And concept seven is the idea of what we call local hosted. And local hosted just means that not only is Hermes running your laptop, but the actual model itself is running on your own computer. Now the trade off for this is that the biggest models are billions and billions of parameters, which means you need data centers that are huge to run them.
14:34So you're gonna be limited based on that, but what you can do on your MacBook, for example, is click the Apple icon at the top, click on about this Mac, and you can see some information about it. So you can basically screenshot that Mac. You can push that image into Claude Hermes and say, hey, for example, if I come down here and do this real quick, I screenshot this, I shoot over to Hermes, I can paste the image in and be like, hey, based on my MacBook specifications, what do you think would be the most powerful local model that I could run?
14:59Okay. And what we can do then is actually download that model using something called Olama that lets you download stuff. And you can even do Olama Cloud, which is a free model hosted somewhere.
15:09Or I can download it specifically like Gemma four, Quan 3.6, whatever, and we can access all those things from your website. Which means that whether you're a thousand feet underground, you are flying in the air, you're in a rocket ship to space, you will be able to use your Hermes agent anywhere with no limitations, 100% private because it is it only exists on your physical laptop.
15:31That is the that is the private and local aspect to it. And as you can see, it's come down and it's given us some different ideas like QUAN 32 b is really powerful. That's 32,000,000,000 parameters.
15:40And these models are just getting better and better. And obviously because it's local and private, you will compromise a little bit on performance and maybe speed because you're not on a huge server. But that's effectively what people mean when they say run local and run private.
15:55Now we've covered this one, it leads very nicely onto concept number eight, which essentially is never starting from zero.
16:02And so I want you to imagine Hermes as a friend that never forgets what you tell her. And it will remember things like your birthday, and it could tell you things that you spoke about on specific days. This is what we mean when we talk about Hermes' memory.
16:14And so for example, I could say something like, hey there, man. Give me an example of a question I asked you on May 19. Okay.
16:21So I can actually ask it questions about specific days, and it can recall that and let me know based on the way that this memory works. And as you can see, the question I asked it was something about a famous thing about being liked and a really cool interesting quote about throwing rocks at enemies. So it has within it, the Hermes agent does a couple of different memory systems.
16:39It has a memory dot m d. And by the way, m d is just a file type. So you know how it's like dot p p t x, like PowerPoint or dot whatever word is, dot x.
16:48Dot m d is just markdown file. All it literally is is just text with no special characters. Like that's basically it.
16:55It's plain, it's readable, and it lives here on your laptop. Then you've got an SQL lite full text search of every message in every session. So it's effectively, it can recall everything you've spoken about.
17:05Don't need to worry so much about the complicated mumbo jumbo, but just know that it's checking everything. And then the idea is that Hermes agent will compound quietly. So ideally, if you you need to use it for two weeks for it to start feeling magical.
17:16And of course, you can supercharge this and connect it to other things like Obsidian memory systems. So for example, in the Hermesogenic operating system I have, I've actually connected it to my Obsidian system here, which means that all of my Obsidian memory, and I'll put a link on screen if you haven't learned about Obsidian or Pancone systems, you can check that out.
17:34And it also is connected to all of my Claude memory. So anything I ever talk to Claude code about or any of the apps that I use on my computer, I can actually access now and it will dream for me overnight and come back proactively with suggestions based on questions that I've asked or how I'm using it or what I'm spending my transcriptions, and it pulls together all these different information sources together to give you really specific stuff, to take its memory to a completely interstellar level.
18:00Because if your Hermes agent isn't getting information from stuff you're doing on your computer, Claude code, codecs, how can it possibly give you the correct and accurate advice? Which is why I think the everything system, bringing it all together into one beautiful operating system is really, really important. Absolutely crushes.
18:17And this brings on to the beautiful concept number nine, which is what we call the character bible, also known as the soul that ended. So this is the file that turns your chatbot into your personal assistant.
18:28It's somebody who gets your vibe, your communication style. If you're somebody who says, hey, dude. Hey, bro.
18:34Or however you wanna talk and communicate, it specifically explains that to the Hermes agent. And what you wanna do basically is give it the information. For example, what is the term?
18:43What are your values? What is your style? How do you want that to communicate?
18:47And what I'm gonna do for you down below is put a fullsoul.md template that I pulled together based on so many hours of using this stuff. It covers stuff like your identity, mission goals, know, things about your business, how you want it to address.
18:58You wanna give that to Hermes and think of it as like the character. Like, do you wanna speak to like an Arnold Schwarzenegger type of character or like a Knight Rider type of character? You can give these soul that in these.
19:08And this is what makes your Hermes agent different from everybody else's because it has a unique soul. Unless we specify what that is, how's it gonna what that looks like?
19:18So super important, give it a soul and make it fantastical. And again, if you're using an operating system, can actually also see what those souls are in your own database. The concept 10 is the idea that Home is itself comes off the shelf with many different integrations already locked in there.
19:33And then what we would basically gonna do here guys is just give it the specific API key for that thing. That could be text to speech, image generation, vision, web search. Effectively, you can connect your entire world to Hermes agent and think of Hermes as a Swiss army knife.
19:48For example, if I come back over here and I open up my Hermes agent, I might say something like, hey there, dude. What was the title of the last meeting I was in and what is one action that I said I was going to take away? So for example, I've connected it to granola.
20:02Not a sponsored video or anything like that, but basically, I use this to capture information on my meetings. It sits in there and basically now I can chat to this because I've configured it within my Hermes agent, and it will now allow me to ask any questions about anything. I could say, hey, dude, what did I talk about six weeks ago for this thing to happen?
20:19Now as you can see, it's gonna come down here and ask my permission to run it. So we're gonna say, yeah, that's absolutely fine. We're gonna allow it for this session.
20:25And what's really cool here guys is actually come back and said, hey, your your connection's expired, which is fantastic. I'm gonna show you exactly how you can add this in for any connection within Hermes because this is a great teaching moment. Now check this out.
20:36Hey, dude. I would like you to give me a terminal command so that I can add my granola API to your environmental variables.
20:45Oh my gosh, Jack. What does that mean? Well, effectively, what we never wanna do is publish our API keys in the chat with Hermes agent.
20:53Because again, everything is indexed and saved. So what we do, especially if it's running on our computer, is we save it somewhere on the computer that it can access, but never actually sits in there. Match it this one on.
21:03It's come down here. All I'm gonna do is copy this right now. I'm gonna open up a terminal like so.
21:08Alright. Here's the terminal. Let me exit out of this process.
21:10Maybe create a brand new one. Okay. Beautiful.
21:12And what I'm gonna do is just paste this one down here and all it's gonna ask me to do is change the API key, which is gonna be right here. Then I come over to my app, I click on API keys, and then once you've created it, I've got this even simplified version here, and then you just literally replace this information here with your API key.
21:27And you're gonna copy that, come to the terminal, and paste it in. That's done. I can say, awesome.
21:32I've updated that. Please go ahead and try it again. And I wanted to include that in just so you get a sense of how you securely and privately update these, again, API keys within your system, giving Hermes the ability to connect to anything.
21:44And anything that has an API key, you can connect to Hermes in exactly the same way. And what's cool about this is Granola also has an OAuth. So if you just give it the name of the Granola MCP like this, it will actually ask you to authenticate and sign in.
21:57And look at this guys, Hermes agent would like to access your account. And the Hermes agent actually opened up this m c m c p server for me. So I come down and I allow access.
22:07And you can see authorization successful. You can now return to the CLI. And all I did guys, honestly, I just came down, I copied this page, I'll just send the link and shared it straight to Hermes.
22:15Now as you can see, it's come down here and it's confirmed that it's found the last meeting, which is a community call I had and exactly one of questions that we asked. And that's and it's such a wonderful example of how you can connect to Hermes agent to your entire world so I could be anywhere and ask it any question I like.
22:30Now concept 11 is the fact that Hermes lives on your computer if you've set it up this way, which means that not only can it do things like create files on your desktop or organize things on your desktop, It can also take actions on your computer. So for example, it can do what? It can do browser, vision, bash, fused, what the hell does that mean?
22:48It has native Chrome Dev Tools, visual screen reading, real cursor movements, not a screenshot bar, an actual agency. So if you're running to the gym and you're like, damn, I completely forgot this thing. Basically, Hermes agent can go ahead and do that.
23:01Now to be fair, this isn't a feature I tend to use a lot. But again, if you just think about the fact that the agent lives on the computer, it can take actions and do various different things for you once you enable it within the tools. Concept 12 is around by using MCPs.
23:14MCPs stand for model contact protocol. So think about it like this. Okay.
23:19You have an a an API is like the wiring and the MCP is like the instruction manual around that wiring. So instead of having like a like, you know, a remote for example, load of exposed wires, MCPs are basically like instruction packages.
23:31It's like a universal remote control that lets you connect Hermes to effectively anything. We just use the granola MCP to connect to basically my meetings.
23:42And we can do the same thing for example with Zapier m c p. So Zapier m c p lets you spin up these servers like so, new m c p server, and I can go ahead and I can connect to anything that I want to. Right?
23:52If come down here for example, let's say that I wanna come down and do other, I come over here and I connect to things that, you know, through this you can't connect to other locations. I can add tools. Say for example, I wanted to give it the ability to read emails.
24:04Right? Well, how would I do that? Well, we can just basically select all the stuff that you wanted to do.
24:07The one thing I wouldn't give it access to is the ability to send emails just yet. I would just hold off on that and its ability to reply. And when that's done, I can add these 10 tools.
24:16I could add in things like Google Drive. I can add in things like whatever I want to with Zapier. It's really cool.
24:21Then all I do, I've got skills over here as a second tab. But what we're really gonna wanna focus on here is the connection. And basically what we can do now is generate a token here and I've got an MCP server to connect to Zapier, which is awesome.
24:33Now what I can do is come back over to Hermes and I can say, hey there, I'd like to connect to the Zapier MCP server. I've got my tokens. Could you give me a command to run to the terminal so I can add these to your environmental variables?
24:45Again, you can post it, but I I'd often recommend that you do this in the terminal so it doesn't get recorded in your chat logs. And once you've done that, you can say, awesome. I've done.
24:53Could you tell me the title of the last email I received? Again, then it will use that MCP to go ahead and do it. Now, why would you create an MCP versus just using an API key?
25:02It is way more what we call token efficient because all the checking it needs to do builds up the cost. An MCP basically explains exactly how it works.
25:12So AIs can basically just look at the MCP manual and know exactly all the stuffs it can do, all the buttons it can press in that individual application. So it knows how to draft emails, create emails. It has full understanding because it standardizes the use of that incredible tool for the agent.
25:30And just like that, we've got all the information that we need from this MCP. Concept 13 is understanding muscle memory for the agent. The idea here is that the longer we use Hermes agent, the better it gets.
25:40What I mean by that is consistently give it feedback as you go so it can iterate and improve as you're doing it. It has a series of Hermes skills that can pick up the more that you use it.
25:51It has various different things it can do. And you can even if you want to build out different skills quite easily.
25:56So for example, you can get to this website here, which is agentskills.im. This gives you a good overview of what skills are if you wanna go deeper on that. One of the things that I did actually guys is I built out something called the Pantheon, which effectively enables me say for example, I wanna add a persona.
26:09Right? I can choose let's go with the Alchemist for instance. I can explain this job.
26:14I can explain this description, give it a system prompt, and then actually choose the model that I want to deal with. Because again, if it's like very intellectualized, we don't want Albert Einstein, you know, washing our car, for example.
26:25You know, we want him working on the boat most impactful tasks. And so you can build out skills like this. Like, I have a deep reasoning task I use for Claude, But, you know, for autopilot jobs, I might use a lower cost model.
26:36So you can build out these skills, which give it the ability to do something different. So for example, I could say something like, hey, that dude. I'd like to do some research for me, some very deep research on the best country in the world to live.
26:47You know, spin up a sub agent. Why don't you go ahead and use DeepSeek v four and chat GPT to critically analyze it in a nice little loop with Opus 4.7 reviewing everything.
26:58Just just go ahead, use a cheap powerful model to go ahead and do loads of deep research, take some time with it, and then come back and share the results based on everything you know about me. Now what's cool about this is it can delegate to different models to do different things for me, and it can use that as a skill.
27:12Now it will use here for me the deep research skill that I've already created, which is fantastic and it just enables you to get so much more out of Hermes. And if anytime you wanna stop it from doing a task, you can do forward slash, stop, and that will actually basically stop Hermes from doing the thing.
27:26The concept 14 is about six keys to make it so unbelievably powerful. So here we've got we've got q. Now what's interesting thing about q here, okay.
27:34Now, q is this, which means that let's say I ask a question like, hey dude, find out for me the best country for octogenarians.
27:42Which country in the world has the highest number of octogenarians? Okay. Really cool.
27:46I can send it off. Then let's say I wanna do a follow-up task, I can do forward slash q. And what this will do is send the prompt on the next terms so it doesn't interrupt.
27:55I can also do forward slash background. Okay. And what forward slash background lets me do is effectively ask it let it do different tasks in the background.
28:03So for example, I could come down and do background and then I might come back and say something like, hey, find out which country in the world has the healthiest 35 year olds or 32 year olds. Okay? Now this will run-in the background because normally when you use Hermes agent, basically what's happening is it's it's interrupting.
28:21So this enables us in one chat window to have met multiple streams of thoughts running concurrently by using those two tasks. And as you can see, background task has gone ahead and actually physically got started.
28:32You got Kanban, which is where a task board functionality can have. Reset will clear it all up. Compress will compress the memory.
28:40So important thing to bear in mind on memory is that there is a certain size for the window. So if you have, say for example, 10,000 words and you ask it a question, every time you ask that chat agent a question, it is literally using all 10,000 words plus your question. And the longer it gets, basically, the lower its performance gets.
28:58So what I do is one session, one window then you want to reset. So compressing basically means, dude, could you just do me a favor and just summarize everything we talked about?
29:07So it goes from this to like this basically, and then you can continue the conversation. And as I showed you earlier, we can switch to models super easily and why don't we come down here and do forward slash model. And then we can basically, now we're talking to Opus, we're talking to ChatGPT, we can talk to anyone that we physically like.
29:23And I put a lot of the interesting ones down here that I thought were really quite interesting that you may find yourself using, you know, basically depending upon what situation you're in the moment. Now concept 15 is what we call basically safety and security for the agent. These are house rules.
29:36Now these agents have access to many different things. And because of that, I think it was Peter Parker who told us this. Right?
29:41With great power comes great responsibility. Mine had been Ben, his uncle. Remember, with great power comes great responsibility.
29:50But the point here is we basically don't wanna give it the keys to the kingdom because that's responsible use, which is why I don't let it have the ability to send the emails yet. A lot of my buddies have done that. They've had their hands burned in the process, And it will get to the point where it will have less errors in humans.
30:05But at the moment, we wanna follow what's called the principle of least access. And what that affects me is Hermes and any agent or system or anything you're building ever, I do this with my clients, you only ever give it the minimal amount of access that it actually needs.
30:19The very, very minimal thing. So for example, there are certain things that we don't want it to go ahead and do. API keys never ever ever post into any chat like a Hermes, especially specifically if it's keeping it.
30:31It's not that Hermes is bad per se. It's just that if you're backing up your Hermes on a daily basis, we just don't want these keys. It's just one less thing for you to think about.
30:39So if you can actually just think about using it responsibly like that and follow this principle of least access, you will generally across all of your AI systems be extremely same. Now the next thing to think about here in concept 16 is called the North Star and I this is what's called the goal function. So think about like this.
30:55Okay? Rather than going question, answer, question, answer. The idea of the goal functionality within Hermes is that you give it a North Star for a short term project.
31:03So I could say, hey, I want to create a YouTube thumbnail that is pretty freaking epic. So for example, if I came over to Hermes, right, and I come down here and why don't I just go ahead and clear this entire session.
31:14And so for example, you can come to Hermes agent and you can give it a prompt like this, which is gone. I would like to decide whether I should take an opportunity a or b. Should do a big consultant role at Anthropic or do I become head of innovation at OpenAI?
31:26I would like you to help me out, help me make the best decision. And look at this, now it's come and said, look, I'm gonna assign a 20 goal budget. I thought a 20 turn budget to this.
31:35And effectively, what will happen now is our entire conversation exists to hit and crush that specific goal. So it will ask me questions about what's it about, what's going on, and it will get there on a midterm basis.
31:46Now what's really cool here is Hermes will not stop until it hits that decision, but it does run into one limitation and that's the fact that not all goals we do in life can be solved in 20 turns and sometimes we've got to do things. This is why I built this idea of super goals, and I'll put a link on screen if you wanted to look into this.
32:03An idea of super goals is that you explain the goal you want to, you use this super goal skill, and effectively what happens here is homies comes and ask you loads of questions. And the output of that is like say four to 10 actions. Some that you do, some that Hermes does.
32:18For example, on this one which is to launch a Hermes course and get 500 sign ups. One is shape the course promise. Build an opt in engine, write a course sequence, then it's my turn.
32:26I have to go ahead and record the video promo and it lets me mark this task as complete when it's done. And then my progress bar goes up. But if I were to remove this and start from scratch, you'll see right now, look, every hearing needs a great goal.
32:38I can copy this prompt. I can shoot straight back over to Hermes. And what I can do is forward slash and I just come down here and click on stop as a for instance.
32:46Say goal pause, what we wanna do. And the command I believe is goal clear. This will end the goal.
32:51And what I can do is come down, give it this information here, and effect from now what this does will allow me to have a medium term goal. And I'll say, hey, Jack. What is the goal that you wanna go ahead and set?
32:59So I'll say, hey, dude. I would like to launch a paid course and get 1,000 email sign ups to my new Hermes Masterclass.
33:06Or whatever the thing is that you want to sell or or help people with. And then it starts to ask questions. So I'll say, hey, in four weeks.
33:13Okay? And then it asks you questions dynamically, and I test this so much, which is why I built all these different specific rules into it, and it's all connected to the dashboard. So when this is done, it will literally load up in the dashboard dynamically.
33:24And again, ask you different questions to get context because that context, how can it know that it's helping you out? Then all you ever do is you land on your mission control, you can manage it in Hermes, or if you wanted to, you can come down and use Claude and it's the exact same thing so you can pick up whenever you want to.
33:39Honestly, these midterm goals, these chief Wiggum loops are so freaking powerful. You can plan your life, hit these midterm goals using Homies Agent. But concept 17 is about using Homies Agent as a team.
33:50So idea here is that we can tag in different models, do various different things. You can spin up what is called sub agents. So for example, if I come back overnight to Hermes agent, what I might do is just say, hey, I'm just gonna pause that for a second.
34:01What I would like to do is go ahead and research for me the best AI companies to work for on the planet. I'd like you to spin up two sub agents. One is gonna go ahead and look at the ones in America, and the other one is gonna look at alternative jobs maybe internationally.
34:15Okay? And what Hermes can do is spin up sub agents. And what sub agents are is think about like this, you're at Hermes, underneath Hermes, you have two other agents each with their own fresh context.
34:27And it delegates these tasks, these two agents go ahead, they do all the work and they report back. Now this is happening in parallel, not sequentially, which means that I could have four, five, six different agents all researching things like a team doing work individually and at the end of the day or the end of the turn, everyone hands in their paper their research, and instead of waiting, you know, six hours, it's done in one hour instead.
34:49This is called parallelization, parallel agents, and gets you significantly further faster using Hermes.
34:56I use this strategy all the time to get the most out of the agents. For example here, Hermes co founder Technum, 12 parallel Hermes agent enters, I hope I pronounce his name correctly about, every day to build Hermes itself.
35:07Monitoring issues, dog food in the Kanban board, Parallel Sub Agents are the power user move. So even the guy that built Hermes agent used Parallel Agents to do it. So getting multiple agents working together in tandem to one objective.
35:19Or it could be individual objectives. Right? One could be doing research, one could be writing, one designer, one scheduler, and they all put these beautiful folders inside your desk.
35:28Now concept 18 is so powerful. This is what's called the heartbeat or the cron job. Cron job sounds so freaking complicated.
35:35One of the coolest things you can do is this idea that it keeps working whilst you sleep. The twenty four seven AI agent. Okay?
35:41A heartbeat effectively is something that you set up to happen every single day. It's essentially called a zombie as well as zombie process if you like. And effectively what it will do is you give it something to do.
35:52So effectively, a heartbeat is silent until something dies. So a sub agent pings every few seconds, zombie detection, fresh agent reclaims a task, per task retries with hallucination recovery. What does that mean?
36:03Let me show you what a cron is first and then I think it'll make a little more sense. So a cron agent here, if I come down here and just say, hey dude, I want you to do two things for me. I want you to send me a reminder in thirty seconds that the sky is blue.
36:15Okay? This is a scheduled task. This is basically in a certain amount of time, it is gonna come and give me an update.
36:22Now for example, I have a daily morning briefing. Okay? And I say I want you to go through my email list.
36:27I want you to go through my calendar. I want you to go through everything I'm doing that day. Use all the information that you know about me and give me, you know, a daily morning brief of say, I don't know, five things that are relevant and it can build up for you.
36:40So every day you get that on repeat. Basically, so you can schedule these wonderful things inside Hermes agent. It gets to come to you at 8AM every single morning.
36:48But the one that you're gonna find yourself using most of the time are these scheduled tasks that make it feel like it's alive because it's proactively coming to you and asking you questions. You could say, hey, periodic times randomly come and ask me questions to check-in on me and do various different things.
37:02Then we're to concept number 19, which is essentially the dirty secret non logic we wanna talk about, which is, can you believe 73% of every request is fixed overhead? So what we're talking about here is budget.
37:13Okay? Now what I mean by that is that, like, if you are using API keys, you can very easily burn through a lot of money, which is why we wanna be very specific about the specific models we're using. Now tokens, you know, a parking meter takes maybe dotted coins.
37:28AI models take tokens. Roughly speaking, a token, I guess a 10 tokens is about seven words. It's about seven seven 75%.
37:37That's effectively what the translation looks like. So just make sure that you often clear the session and that you're using the right model for the right job when you're using Hermes to keep your token cost usage all the way down. And there's loads of many strategies like don't overload it with a bazillion skills that you don't need.
37:53Don't make it system prompts too long. Keep it nice and clean. And one conversation, one goal, then it's always best practice to clear and refresh and start again.
38:02And just to point this out, some guy actually spent 4,000,000 tokens in two hours of light usage, 21,000 tokens just asking about the weather because he made a mistake. So just give you a good a good idea of how you can reduce those down.
38:15Now the twentieth concept here is this idea, but it isn't just a chatbot. Hermes agent can be part of an operating system where we have various different personas in the pantheon.
38:26We can have a memory vault that extends its memory capabilities to a significant level, basically enabling you to call upon any email or meeting that you've ever had, any knowledge source from any YouTuber you've ever seen connected to Notebook alone, a gargantillion different things.
38:43We can have, for example, so many different tools. And effectively, when we create an operating system, it just gives it a place to manage everything. I call this a kind of everything of AI.
38:52So if you look at this called code operating system, we have our AI spendage, how much money we're making from our skills, we can see our usage in real time, dreaming improvements. Like I said, that will come to you proactively, mission control our skills, and memory systems.
39:05Now in the Hermes operating system here, I can see with a glance all of the connections that it has available to us. So I know exactly what it's connected to. I know what model I'm using.
39:14I know my memory. I know my usage. I can chat to it directly.
39:18I have this ability to manage goals and plans with Hermes agent. So whatever I'm working on, I can work with Homeless Agent in tandem on that. I have this Pantheon, so I can create these beautiful I just personally like to visually see my graphics and what I'm dealing with.
39:31So I can like, well, actually, no, I really wanna kind of upgrade the okay. I'm I'm missing this or missing that. Can grab different things and build various different things.
39:37I just think it's a nicer user interface to be able to go ahead and build things like that. Again, and what we can do here crucially, and I highly recommend you do, is connect your Hermes to GitHub. Now GitHub is basically code for place where we store stuff.
39:51Okay. It's like the world's repository. It's basically we can store information files.
39:55And what you can do with Hermes agent guys is essentially have your entire Hermes agents backed up on GitHub every single day. Mine does that. And what that means is if you ever change a computer or God forbid you lose it or something, you have that backed up somewhere.
40:09Right? We can connect to GitHub. We've got all the profile information here.
40:12We can connect to our Obsidian brand, which is this beautiful memory system here that basically just lets us see everything we're doing with different memory systems. We've got that as my Cloud.
40:20M d and just enables you to get a level of control. And crucially comes on to the twenty first concept, which is the idea of one brand.
40:29The idea that we are going to connect the Hermes agent, the business brand with ColdCode. We talked about the differences, but here's the difference guys. If we can actually connect them together, it unlocks incredible capabilities.
40:41Hermes knows your projects, your customers, your decisions. It lives everywhere. It is everywhere.
40:45Maybe it's like, and who knows what it And then you have Claude Code is this precision tool that we use for building. And actually what happens is when we can physically connect these ones together, we have this shared memory and understanding where we're not operating in isolation.
40:59It's like you say one thing to Hermes agent, another to Claude Code, how do you know they connect together in a meaningful way unless they have access to basically what they're doing at different times. And so now we've covered these 21 core concepts in plain English. The next thing we need to do is leverage all of these and get one of these operating systems ourselves, which you can learn about in this video right here.
The Hook

The bait, then the rug-pull.

Most AI tutorials assume you already know what an agent is. This one starts before that — at the moment you open the app, see a wall of tools and skills, and ask yourself what any of it is supposed to do for you.

Frameworks

Named ideas worth stealing.

01:21model

Four Different Animals

  1. Hermes — your dog (lives with you, compounds over time)
  2. Claude Code — your contractor (brilliant, project-scoped)
  3. OpenClaw — your roommate (always around)
  4. Antigravity — your IDE buddy

A mental model for when to use which AI tool based on the relationship each requires.

Steal forAny explainer comparing multiple AI tools
05:48list

Where Hermes Sleeps

  1. Your laptop — FREE
  2. VPS — ~$5/month
  3. Serverless — cents/month

Start local and free; only pay for always-on infrastructure when uptime genuinely matters.

Steal forAny self-hosting explainer
09:12model

Two Doors to the Brain

  1. OAuth — one button, revocable
  2. API key — secret string, rotate when exposed

The only two ways to connect any service to Hermes.

Steal forAny API explainer for non-technical audiences
11:33model

Pick The Right Tool Toolbox

  1. Claude / Opus — reasoning
  2. GPT — generalist volume
  3. Grok — volume + Twitter search
  4. DeepSeek — free, high-volume

Model-agnostic strategy: treat your AI brain as a toolbox, not a single hammer.

Steal forAny content about AI model selection
27:28list

Six Keys — Shortcuts That Unlock It

  1. /q — queue next prompt without interrupting
  2. /background — run task in parallel
  3. /kanban — task board view
  4. /reset — clear session
  5. /compress — summarize context to shrink it
  6. /model — swap the AI brain mid-session

The six slash commands that separate manual Hermes users from power users.

Steal forAny power-user tips video
31:49list

House Rules For The Agent

  1. ALLOW: git status
  2. BLOCK: rm -rf
  3. ASK FIRST: send email
  4. REDACT: API keys

The principle of least access applied to AI agents.

Steal forAny AI safety or responsible-use explainer
33:53model

A Team, Not a Bottleneck

  1. Researcher
  2. Writer
  3. Designer
  4. Scheduler

Parallel sub-agents each with fresh context work simultaneously and report back.

Steal forAny content about AI delegation or productivity
38:03model

Watch The Meter

  1. 73% of every request is fixed overhead
  2. 2-3x more tokens on Telegram vs terminal
  3. 1h cross-session Claude cache window
  4. Cheap model for the loop; expensive model for the hard call

The dirty secret of AI token costs: most of what you pay for is boilerplate, not your actual query.

Steal forAny AI cost optimization video — the 73% stat is surprising and shareable
CTA Breakdown

How they asked for the click.

VERBAL ASK
40:31next-video
“The next thing we need to do is leverage all of these and get one of these operating systems ourselves, which you can learn about in this video right here.”

Clean end-screen redirect to OS setup video. No product pitch, no newsletter. Pure next-step funnel within YouTube.

MENTIONED ON CAMERA
01:21productHermes AI Agent (open source) ↗
11:33toolOpenRouter ↗
14:20toolOllama ↗
19:34productGranola (meeting notes) ↗
23:13toolZapier MCP ↗
16:02toolObsidian ↗
25:36linkagentskills.im ↗
38:19toolGitHub ↗
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Storyboard

Visual structure at a glance.

open
hookopen00:00
stats slide
promisestats slide01:07
four animals
frameworkfour animals02:49
one brain 22 mouths
frameworkone brain 22 mouths05:36
where hermes sleeps
frameworkwhere hermes sleeps06:20
two doors to brain
frameworktwo doors to brain09:27
pick right tool
frameworkpick right tool11:35
never starts from zero
valuenever starts from zero16:02
character bible
valuecharacter bible18:17
every blade included
valueevery blade included19:26
it becomes the hands
valueit becomes the hands22:39
every app one port
frameworkevery app one port23:11
muscle memory
valuemuscle memory25:33
six keys power tool
valuesix keys power tool27:26
house rules
valuehouse rules30:30
team not bottleneck
valueteam not bottleneck33:56
it keeps working
valueit keeps working35:28
watch the meter
valuewatch the meter37:18
not a chatbot OS
valuenot a chatbot OS38:16
connect to claude
ctaconnect to claude40:39
final CTA
ctafinal CTA41:09
Frame Gallery

Visual moments.

open00:01Frame at 00:30 from Every Hermes Concept Explained for Normal People00:30stats slide01:07Frame at 01:59 from Every Hermes Concept Explained for Normal People01:59Frame at 02:25 from Every Hermes Concept Explained for Normal People02:25four animals02:49
open00:01Frame at 00:30 from Every Hermes Concept Explained for Normal People00:30stats slide01:07Frame at 01:59 from Every Hermes Concept Explained for Normal People01:59Frame at 02:25 from Every Hermes Concept Explained for Normal People02:25four animals02:49Frame at 03:41 from Every Hermes Concept Explained for Normal People03:41Frame at 04:36 from Every Hermes Concept Explained for Normal People04:36Frame at 05:08 from Every Hermes Concept Explained for Normal People05:08one brain 22 mouths05:36where hermes sleeps06:20Frame at 07:04 from Every Hermes Concept Explained for Normal People07:04Frame at 07:55 from Every Hermes Concept Explained for Normal People07:55Frame at 08:26 from Every Hermes Concept Explained for Normal People08:26two doors to brain09:27Frame at 09:55 from Every Hermes Concept Explained for Normal People09:55Frame at 10:19 from Every Hermes Concept Explained for Normal People10:19Frame at 10:38 from Every Hermes Concept Explained for Normal People10:38Frame at 11:18 from Every Hermes Concept Explained for Normal People11:18pick right tool11:35Frame at 12:27 from Every Hermes Concept Explained for Normal People12:27Frame at 13:11 from Every Hermes Concept Explained for Normal People13:11Frame at 13:29 from Every Hermes Concept Explained for Normal People13:29Frame at 13:55 from Every Hermes Concept Explained for Normal People13:55Frame at 14:51 from Every Hermes Concept Explained for Normal People14:51Frame at 15:03 from Every Hermes Concept Explained for Normal People15:03Frame at 15:45 from Every Hermes Concept Explained for Normal People15:45never starts from zero16:02Frame at 16:36 from Every Hermes Concept Explained for Normal People16:36Frame at 17:18 from Every Hermes Concept Explained for Normal People17:18Frame at 18:11 from Every Hermes Concept Explained for Normal People18:11character bible18:17Frame at 18:45 from Every Hermes Concept Explained for Normal People18:45every blade included19:26Frame at 19:50 from Every Hermes Concept Explained for Normal People19:50Frame at 20:25 from Every Hermes Concept Explained for Normal People20:25Frame at 21:21 from Every Hermes Concept Explained for Normal People21:21Frame at 21:49 from Every Hermes Concept Explained for Normal People21:49Frame at 22:15 from Every Hermes Concept Explained for Normal People22:15it becomes the hands22:39Frame at 22:53 from Every Hermes Concept Explained for Normal People22:53every app one port23:11Frame at 23:27 from Every Hermes Concept Explained for Normal People23:27Frame at 23:43 from Every Hermes Concept Explained for Normal People23:43Frame at 24:35 from Every Hermes Concept Explained for Normal People24:35Frame at 24:53 from Every Hermes Concept Explained for Normal People24:53muscle memory25:33Frame at 25:56 from Every Hermes Concept Explained for Normal People25:56Frame at 26:04 from Every Hermes Concept Explained for Normal People26:04Frame at 26:34 from Every Hermes Concept Explained for Normal People26:34six keys power tool27:26Frame at 27:54 from Every Hermes Concept Explained for Normal People27:54Frame at 28:46 from Every Hermes Concept Explained for Normal People28:46Frame at 29:14 from Every Hermes Concept Explained for Normal People29:14house rules30:30Frame at 31:22 from Every Hermes Concept Explained for Normal People31:22Frame at 31:59 from Every Hermes Concept Explained for Normal People31:59Frame at 32:19 from Every Hermes Concept Explained for Normal People32:19Frame at 32:41 from Every Hermes Concept Explained for Normal People32:41Frame at 32:59 from Every Hermes Concept Explained for Normal People32:59Frame at 33:30 from Every Hermes Concept Explained for Normal People33:30Frame at 33:39 from Every Hermes Concept Explained for Normal People33:39team not bottleneck33:56Frame at 34:59 from Every Hermes Concept Explained for Normal People34:59it keeps working35:28Frame at 35:49 from Every Hermes Concept Explained for Normal People35:49Frame at 36:19 from Every Hermes Concept Explained for Normal People36:19Frame at 36:48 from Every Hermes Concept Explained for Normal People36:48watch the meter37:18Frame at 37:39 from Every Hermes Concept Explained for Normal People37:39Frame at 38:05 from Every Hermes Concept Explained for Normal People38:05Frame at 38:11 from Every Hermes Concept Explained for Normal People38:11not a chatbot OS38:16Frame at 38:35 from Every Hermes Concept Explained for Normal People38:35Frame at 39:05 from Every Hermes Concept Explained for Normal People39:05Frame at 39:48 from Every Hermes Concept Explained for Normal People39:48Frame at 40:00 from Every Hermes Concept Explained for Normal People40:00connect to claude40:39Frame at 40:56 from Every Hermes Concept Explained for Normal People40:56final CTA41:09
Watch next

More from this channel + related breakdowns.

31:08
Jack Roberts · Tutorial

Hermes Agent just got 10X Better (Agentic OS)

A 31-minute setup walkthrough that bridges Hermes AI agent and Claude Code into one shared operating system — with Pantheon personas, Obsidian memory, Apollo lead scraping, and Zapier-to-Gmail wired in by the end.

May 15th
Chat about this