How to Build Your Own Agent Operating System

**Julian Goldie** (0:00)
Today, we're going to talk about how to create an agent operating system. If you're wondering what an agent operating system is, this is, for example, all our agents working together. We've got Claude, Hermes, OpenClaude. We can build out our own SaaS tools with these agents as well. So for example, we built our lead generation and outreach tool using Hermes. We have the Hermes Oracle that can find the latest news and basically automate anything that we want from here. We can also, for example, have custom builds for video agents or SEO, or even music itself, right? And so what we've done here is build probably the most powerful system I've ever seen with AI. And we've plugged it into one system called the Agent OS. And then with the Agent OS, we can basically keep improving it every single day. It becomes a system that improves every single day. And what I'm going to show you today is how to build it, because I built this from scratch and literally have no experience building SaaS tools. I have no experience building this stuff out. I have never done this before. But literally from, I would say, about eight weeks ago, I started this project and it's become an absolute beast. It's become the most exciting thing that I've ever worked on when it comes to AI. And it's become more powerful than anything else I've seen out there. So if you want to build something similar, I want to show you how you can build it using a five layer system, so you can build your own agent operating system one layer at a time. And most people are running like AI's 14 scattered tabs all over the place. Minus one operating system stacked in five layers. And we'll talk about those layers in a second. We talk about foundation, memory, routers, agents, and then the loop. And here's exactly how to build your own mostly for free if you want to. I mean, because I already have the Claude subscription for Claude code, I don't pay anything extra to build this stuff out. Right. Like it just come. It's just basically I use the tokens I've already got included in my plan. And then we just keep building out more and better stuff inside it. And it becomes the ultimate tool that we can build an ultimate anything with. And if we create too much, then we can just hide it inside the settings. If you want to share it, we can. We actually shared it with our community. So we've got three thousand nine hundred members inside our Air Prof. Boarding community. We've shared the agent operating system with everyone there. And then we've got this whole memory system, as you can see, that basically links everything together so that all of our agents have context. So it works in five layers. Let me talk you through each section. This is the mission control, as you can see right here, the actual mission control dashboard. And the way that I look at this is like every layer is built on tools you probably already have. So for example, I built it with Claude code. I've got a Lambo in there. You can use Obsidian. You've got, for example, GLM 5.2 plugged in as well into our agents. We have a router to decide which model wins. And I'll talk more about that in a second. And then we have a full agent operating system. And the main thing to note here is like, you know, I ran AI previously as 14 different tabs, and then I built it into an agent OS. I had chat, GPT in one tab, Claude in another, a coding tool in the third. Super messy. Every session I had to re-explain my business.
I had to explain my clients, my voice from scratch. Files got lost, contents got lost, nothing got remembered, and everything, you know, pretty much became a mess from there.
That was the old way of doing things, right?
Now, we have a way where I stopped collecting tools and started stacking layers. So now it's one agent operating system with every model and every agent on one screen. They share one memory, so they know my business literally as soon as I start them up. So if I go to Hermes, I'm like, tell me what you know about my business. So tell me what we worked on last week. It would give me a day by day recount because it understands everything about me, my business, what I work on, etc. And so it's one agent operating system for every model, an agent on one screen. They share one memory, so they know my business. Each job goes to whichever model wins it, mostly free ones. And the agents do the real work 24-7, whilst I go off and do everything in the background, right? And also the great thing about this is we created a project, which we've shared with everyone inside our Air Profit Board and community. And at this point, we've got 194 pages of wins and testimonials from people using this stuff too. So look at what Shin said, right? He said, out of all the school channels have joined, the agent OS system is by far beyond the worth the value. Right. So Hokey said, I've installed the agent OS and your system is amazing. Rick said he joined the school group and he's already got superpowers and that he built it in just 30 minutes. You got, for example, so many people just sharing their wins, sharing what they've built with it, sharing their customizations. Like, for example, Sheena, she built a self running lead machine using the agent OS as well. So you can customize it and build it however you want as well, which is a fantastic thing as well. So if you're watching this, I promise yourself that you'll set something like this up because the sooner you do, the more fun it's going to be, the more powerful it will be. And the sooner you will enjoy building with AI instead of feeling overwhelmed by it. That's the thing that I would say is like, it makes it so much easier to control when you build the system yourself. Whereas, for example, if you're chasing every single model, every update, every new tool that comes out, it's going to be super messy and overwhelming. You're going to have shiny object syndrome. I can imagine a lot of people watching this right now feel overwhelmed, feel like there's too much going on. So we've basically set up in five layers. So let me talk about how this works step by step. And also, I want to say here, like the old ways, you have 14 tabs, you have loads of different tabs everywhere. You re-explain yourself over and over again. You lose all your files and all your contacts. You're using all these different tools monthly. It's very hard to keep track of them. Nothing talks to anything. No memory, no handoff. And you are the glue holding all of your AI tools together by hand. That's simply what it comes down to for a lot of people who are watching this sort of stuff. Whereas the new way with this system, with the Agent OS, is like every model and agent is on one screen. You got a shared memory, so it knows your business. Each job is rooted to the model that wins it. Agents ship real work whilst you sleep. It compounds and gets stronger every week. And the system is a glue. You just point it and that's the difference. The system is a glue. It's not you just trying to juggle a million different tabs of AI tools all together. So the first step really is the foundation. And this is every model inside one screen. So if you look on our side panel here, what I recommend you do first of all, is every tool you actually use, not every tool that you have, but every tool you actually use, plug into the agent section in. Have a sidebar like this. So layer one is really the ground that everything stands on. One dashboard with every model wired in, not 14 tabs in one place. And you can have a free local model for the everyday work. You can have free APIs for more. You can have the CLIs you already pay for, including Claude plugged in as well, so you're not paying twice. And then you've got everything inside here ready to go. So if we want to speak to Claude, no problem. We've got it down here. If we want to speak to Hermes, no problem. We've got it down here, right? That's the point of this whole system. Now, if you're wondering, how do you get started with this? You just go into Claude and you tell it what you want. So if you're like, right, I want a mission control dashboard that looks beautiful, with Claude and Hermes built in, you just tell it what you want. If you want my full setup, you can get that inside the app off the board. But if you want to build it from scratch, you can. So you might say, okay, well, doesn't this use a lot of tokens when you're building out an agent OS system? No. That's the biggest myth about it because the foundation can run stuff on three local models. It can run stuff on three APIs and that way you don't need to worry about tokens at all. Now, layer number two is the memory. And the memory is really just one shared brain. Now we use Obsidian for this. As you can see right here, this is our memory galaxy. It's improving and growing. The universe is expanding every day, my friends, right? And so it's one shared brain. And the thing that I would say here is like a model with no memory is a stranger. You reintroduce yourself to every single morning. So layer number two in this whole setup fixes that because you got one shared vault, your business, your clients, your voice, that every model and agent reads before answers inside the system because they're all linked. And so you stop wasting 20 percent of every prompt, re-explaining who you are, already knows. And this is your Obsidian vault, Hermes, my video agent, my SEO agent, Claude, etc. They all plug into this beautiful vault. And then one vault feeds context to every agent. So they all know you and your business for the answer.