Agent OS + Obsidian + Memory Galaxy!

Key Takeaways

So, today we're going to be answering some of the latest questions on the agent operating system, which is a powerful system where you can link all your agents together. You can add new updates as they come out. So, for example, we've already added Sakana Fugu that just dropped yesterday and fusion inside our system to achieve Fable 5 level intelligence without having access to the models. We also have a memory galaxy over here. We've set up a new news radar that's just working 24/7 finding us some the latest headlines so we never fall behind again. We have a music agent, a video agent, SEO agents, loop engineering, etc. Everything we need is built inside this system. We've actually put it inside the AI Profit Bot. If you want to get my setup, but whether you're building this or using mine, I'm going to talk through exactly how to use these systems to get the most out of them based on the latest questions we have inside the AI Profit Bot. And so, I'm going to get straight into this and we're going to start going from here. So, we've got a question from Amaterasu, which is about how to keep the agent operating system current. So, for example, let's say you set it up, how do you update it? How do you add the new additions, etc.? So, the I mean, there's two things to this. Number one, when I'm building it, I add new features every day based on what just comes out. And the cool thing about this is you never need to really stay updated on what drops with agent operating systems and AI models because we already built it into the agent OS. But also inside the zip files, every time we update it, we add a new update section so that you can easily update your version. So, if you're on a previous version of our agent operating system, you can just grab the zip file and it has an update file inside it that allows you to update quickly. So, if we go inside this section here, we have an update agent OS command section and that basically allows your agents to stay updated. So, we have a markdown file called update.md and I update agent OS. command file as well to update quickly. And that's one of the biggest benefits of of getting this system is like you can stay updated daily with the new stuff that drops. Uh this is pretty wild. Uh so you can see here that Daniel is generating local AI avatar videos. So if we have a look here, he's actually created a full tutorial on how to do which is pretty insane. If we have a look at the output here, let's see what we've got. So this is the AI avatar video. And I think 99% of people are not aware that you can do this locally if you have a good setup. He has a RTX 5090. I think it's a fantastic example of people sharing information together inside this community as well. And you know, when we work together inside the Air Pro for audio like we all just help each other grow. We help each other learn. There's no holding information back. It's more about creating a a great place where we can share ideas and we can work together and we can grow together. So I mean, it's pretty amazing that you can do that locally. Absolutely awesome. The other thing I would say is actually managed to set up a workflow where it has memory as well. So one thing that we have built into the agent OS is we have a memory galaxy where our agents automatically update this daily. They also use it for context and then your all your agents have context on you. They know who you are, your goals and everything else. And so he's managed to use similar system to give memory to the agency he actually works with. Now this is an interesting question. So Blake is moving from a Claude to Hermes. He said everyone, so I already built my own personal agent system. So he's not actually using ours. He's built his own personal OS system in Claude. Does this migrate to Hermes when he installs Hermes from here and connect to my Claude? How does that work? So in my experience, when you're setting up Hermes and you already have Claude set up, I would go inside Claude and ask it to orchestrate Hermes for you. And that way they can link together and they can share context and also they can migrate everything over together. You can also ask Claude to build Hermes directly into the personal agent OS that you already have and make sure that it has access to the same context and same knowledge, the same skills, etc. And that way they can share stuff. I actually find that Claude is really good at orchestrating Hermes. Hermes is not so good at orchestrating Claude, so I'd do it in that way around. And that's the cool thing about these agent operating systems, right? So, for example, if we have a look in you know, if you look over the last couple of weeks, GLM 5.2, Kimiko 2.7, as soon as these new things drop, we plug it into the agent operating system. And you don't need to use them, you don't need to use all these APIs or CLIs. Like, you can just stick to Claude and Hermes or you can just stick to Hermes if you want. That's the great thing about it. It's like, you know, if you if you have looking at this and thinking, "Oh, I need all these CLIs." You don't. You just you can use free APIs, local models, and you can just stick with something as simple as Hermes, right? And still get a lot of power out of a system like this. So, that's the great thing about it. It's very flexible depending on your setup and what you want to add in and take out. So, I got a question from Aaron here who's looking for thoughts on initial hardware setup. So, looking for some guidance on the best hardware setup. Should you use a PC? Should you use a Mac? Especially if you're traveling 40% of the time. So, what's the best setup, etc. So, for me personally, if I'm traveling, I like to use my MacBook Pro. Personally, I find that it breaks way less than whenever I've used Microsoft. And also, a lot of the new updates, particularly new software or apps, usually comes out for Mac first, which means you get access to the new stuff before everyone else. So, I'd recommend going with that instead. And if you're not running local models, you don't need anything powerful. Just a basic MacBook or MacBook Pro will do the job. Good question from North Campbell who says, you know, is anyone else finding the AI can create a strange illusion of progress? So, sometimes you can spend like all day improving something, and then before you know it, you spent you know, 6 hours optimizing a workflow that doesn't even have a real use case yet. And this is very I've seen loads of people do this. Seen loads of entrepreneurs like spend all day on Claude code and not actually get anything out of it or not actually build anything or not even they kind of neglect the stuff that they should be doing, you know, and that doesn't always involve AI personally. So, the way that I look at this is audit your time, and we've got a training module inside the Alpha Efficiency Room, but audit your time, understand where you're spending your time, and then just focus on automating that. So, never dive straight into building something or automating something before thinking, okay, will this bring me closer to my goals? How much time is this going to save me? And also, is this 100% necessary? Because before you even automate something, you've got to look at that task and think, okay, do I need to automate it or do I just need to eliminate or delegate it first? So, that's the mental framework that I take. This is a pretty interesting system here. So, uh what you can see Dan is talking about is basically, you know, if you're using free models with small context windows, and then you uh building something that requires more tokens than the actual context window, it's got no room to respond, and it just outputs something that's pretty bad. So, how do you fix it? So, one thing that he recommends is estimating tokens before you call the API. And that way, if it's too big, you actually get a warning before you generate the response. And I think that's a fantastic way to do this as well. Really great idea for just managing things efficiently and also getting the most out of this. I actually found this with Sakana when I was testing out. Sometimes it would output something that hit the token limit. And then you'd spend 20 minutes waiting for a response. That actually didn't work at all. So, yeah, it's a great way to look at things. Jose is setting up his agent OS fixed with Hermes. And one really good tip here is that you can use and videos API in Hermes for free. One thing I've seen with that is sometimes you can hit token limits pretty quickly. So, just be aware of that. But yeah, you can get free APIs for large LLMs for free, which is pretty cool. Now, one thing that's interesting recently is the Gemini CLI has actually been replaced by antigravity. That just happened over the last few weeks. So, we're actually removing Gemini CLI from agent OS because you would just use antigravity instead. Now, here's a good question from Amita. She she's setting up the agent OS system as well. And she was like, do you need like a you know, an API key to set up Claude inside your agent operating systems? You actually don't because you can just use the CLI to get access to that. So, all you need to do is just make sure that you sign in. And the instructions are inside the setup guide. You can also use terminal as well. If you prefer, you can use an API key. I also have free Claude code inside there, which allows you to use free Claude code with the agent OS system as well. So, then you can even use like free APIs for the system, which is pretty cool. So, Robert was asking about open router fusion. And basically what this is is a new API where basically you have multiple models as a panel working together. And they can actually outperform Fable 5 on benchmarks. Now, this is something we've already plugged in here. So, you can see the Fusion boardroom here, and Sakana are another AI model that set up the same system. So, we've added both inside the Agent OS. So, if you want to use Fusion, you've got an example right here. And you can build some pretty crazy stuff. I mean, for example, you can see an example of what we've built with it right here. Uh it also built out like a a 3D city simulator, which is pretty amazing. And it can even create like, you know, these full open-world RPGs, which is pretty wild in itself. So, it's been really impressive so far. If we actually have a look on Goldy bench, you can see how it performs on my own benchmarks here and our tests. So, you can see how these models perform step by step. And Fugu, which is an alternative to Fusion, is currently outperforming Fusion. And so, we have a look at some of the stuff that we've built with this. Let's open this up. See what we got. All right. All right. So, this is uh ray caster maze. As you can see, we built this with Fugu. Pretty amazing. A website. This was a best generated website. Much nicer than Opus 4.8 or anything else. Created this insane galaxy simulation, which is absolutely wild. Uh in a solar system emulation, another one here, too. And then this liquid in a bowl simulation, which is pretty cool. And then we've got like a a voxel runner game as well, which is by far the best that we've generated out of everything. So, on every test that I do personally, like Fugu Ultra is is goated. It's absolutely goated. And then that is followed up by Fusion. So, Fusion is pretty cool, too. You can see some of the stuff that we built with it over here. And it can create awesome stuff, too. So, let's take a look at this one, for example. This like a 3D game we created with Fusion. And the outputs of this are insane. So, when I look at the benchmarks here, we we can actually compare them side by side. We've got comparisons for every model that I've personally tested. And this is all based on the stuff I've tested personally. So, you've got Fugu Ultra versus Fusion. And so far, Sakana is outperforming Fusion. But, the the other thing to note here as well is like it takes so long for responses. Like sometimes it can take 20 minutes per generation for Fugu Ultra, which means that obviously we've still got a lot more testing to do. But, side by side, they're pretty powerful stuff. So, Fusion is from OpenRouter. Sakana have done Fugu. And it's pretty awesome. Next up, we have a question from Blake. So, Blake is using Codex to build a website as a knowledge network for the real estate industry. Can he use Hermes to build a robust mission control dashboard and agent operating system? So, you definitely can use Codex. You can also just use our example below and then just tweak that exactly how you want. I'd actually recommend that because then you can use our setup as a base and tweak it and customize it how you want. That's what a lot of our members do. So, I'd recommend that approach and you can easily give this setup to Codex and tweak it and improve it from there. But yeah, Codex is a great tool for building with. We've got a post from Rachel who says, you know, thank you for everything. I think this is great. Like, it's just such a positive community. So many people sharing stuff. So, you can see here, she says, I just want to thank everyone in the community. Your insights, support, practical AI strategies have helped me grow and improve. I'm grateful to be learning alongside people who are actively applying AI and sharing real value every day. This is what we like to see, just a great community of awesome people. Thank you so much Rachel for your feedback. And this is what I love about the community. It's just everyone winning and learning and growing together. And that's what it's all about. And if you're thinking if you've watched this so far and you're thinking like this stuff is super technical, blah blah blah. Actually, you can see we've got over 187 pages of testimonials and wins like this. I mean, Brick set up his Agent OS in 30 minutes from our framework, so you know, Rachel's getting awesome results as well. There's so many great people getting amazing results inside this community that I know if I'm non-technical, maybe you're non-technical if you're watching this. A lot of the people inside the community are non-technical, too. You don't need to be a technical person to use this stuff, right? You just need to focus on, okay, what's the next thing I need to automate? And then how can I do that? And I think the Agent OS helps you really apply that. This is a really good post from Alex as well. So, Alex is talking about his win, you know, moving from AI tools architecture. And this is really the way that I've changed my mindset recently as well. You know, so he's been talking about basically instead of looking for one tool that does everything, design around specialized models, automation workflows, quality control layers. So, the way to think about that is that when you have a system, that is way more important than any model or any framework or any benchmark. Because when you have a system, you can tweak it and mold it exactly how you want. But you can't do that with a model. So, if I go into, for example, like Claude desktop, I can't mold it the way that I want it to be. But I can change this whole system here. So, if I spend a lot of time on videos, I can build a fantastic system around that. If I do a lot of SEO, I can build an automated workflow that does SEO exactly how I want it. And so, the way that I look at this is doesn't matter what Fable can do, doesn't matter what Fusion can do, it doesn't even matter what Sakana Fugue can do. What matters is do I have a great system that actually automates every part of my workflows in a quality controlled way. And that's exactly what we've built with the Agent OS system. So, you want to think in terms of systems over models. When you build a great system, it doesn't matter what models or benchmarks come out because they are interchangeable and they literally change every single day. But when you have a system, that improves daily. And when you improve it every week or every month or every day over a year, the wins add up and you build something amazing. This is a great question. It says, "Anyone actually built something like Jarvis?" You know, like a Jarvis-style page in an Agent Operating System. That's exactly what we did inside the Hermes Jarvis section of the Agent OS. So, basically we have this real-time model that can understand what we're talking about here. And you can see that we are talking to it and it can understand things. So, if I switch off real time, it's switched off. If we set up this wake word button, then it'll wake when I give it a command. We can change the voice over here and we can also get a daily briefing. Plus, we can see the full history of Hermes Jarvis and all the stuff that we've built with it, which is pretty insane. We use ChatGPT real-time and it's all built into our Agent Operating System, which you can get inside the app for a bottom if you want a ready-made setup. So, open up Obsidian and boom, look at that. It just it can use computer use, it can open stuff up, it's all good to go, and it's ready whenever you need it. Check out the Hermes Jarvis section. What it can do is you can speak to it in real time, it responds instantly, you can change the accent on it, you can build stuff and you can preview it. It has everything saved inside the gallery when you do preview it. It also has access to computer use and you can basically operate Hermes through your voice and it's voice activated. Plus, you can see your full conversation history and you can even have it in warm mode, which means that you can have it full screen on a monitor in the background and then you can command it whenever you need to. It also has weekly and daily briefings inside there. And we have a full guide on that, too. This is pretty cool as well. So, if you're using Obsidian as a memory system, which we have over here as you can see, some members are like really customizing it in an absolutely insane way. So, let me give you an example. So, if we look at this, this is Benjamin's Obsidian setup and you can see like the way that it's customized, the way it's organized, it just looks super nice. Like even the feel of it and the vibe of it matches, for example, that philosophical style that he's going for. You know, it's kind of it feels stoic when you read it, you know? And the same for this, for example, the Oracle that he's built as well, which is his voice-activated AI agent. So, really like the customizations, looks great. I like the fact as well like you're making it your own. And also, the style and the vibe of your design matches the exact type of content you're creating inside the Agent OS. Awesome. Tyler is asking about Fable 5's uh similarity with Sakana Fugu. So, as I mentioned before, we've already tested it out. We've already got it built into the Agent OS as well. And it's really good, really good, especially Fugu Ultra. So, it's really good. The only thing is the there are API limits and also it hits the token limits pretty quickly and it can take about 15 to 20 minutes per generation. But, it's really good if you need to like build out something foundational, like a big project or some code, and then you can get your other agents to tweak and improve it from there. You could even for example use it for planning a project, managing all the potential forks and then get your other agents to build out quickly. So, you could create a plan with fusion or Sakana Fugue and then from there your other agents could implement it. That's another option. We've already tested out on Goldy bench and it's top of the leaderboards. So, I've answered every single question inside the community today. As you can see, these are questions that you can ask at any time and also the cool thing about this is for example that I personally answer it and so does the whole community and we're all just raising each other up and growing together. And also the agent operating system that we're building out day to day is just getting better and better as well. Now, inside the classroom you get access to all of our best training. So, you can see for example we have a 6-week beginner to expert training course right here. You can also get our agent operating system which is updated daily with new updates inside this section. So, you get the video tutorial, the last update date so you know it's recent and also the zip file to install it plus a guide on that. And also every time we test something new for example like Sakana, we test it out within an hour of it coming out. Every time something new like that drops, we add new daily advanced tutorials. So, whether you want the beginner stuff or whether you want the advanced stuff, we've got it all inside here along with the systems that make this stuff great. Now, inside the calendar you can jump on weekly coaching calls, get help and support in real time. Inside the map, you can actually meet people in your local area who are building with AI agents as well which is great. And this is all inside the AI Profit Bot Army. Link in the comments and description or just go to the aiprofitbotarmy.com. Thanks so much for watching.