---
title: 'How to Build a Local AI Agent Operating System with Hermes'
source: 'https://youtube.com/watch?v=bZRBThn2xLw'
video_id: 'bZRBThn2xLw'
date: 2026-07-14
duration_sec: 0
---

# How to Build a Local AI Agent Operating System with Hermes

> Source: [How to Build a Local AI Agent Operating System with Hermes](https://youtube.com/watch?v=bZRBThn2xLw)

## Summary

This video demonstrates how to build and use a local AI agent operating system (Agent OS) using Hermes and other tools. The presenter answers community questions, showcases local AI setups for generating images and videos, and explains how to orchestrate teams of AI agents for automation, all while keeping data private and free.

### Key Points

- **Introduction to Hermes Agent OS** [00:00] — The video kicks off by introducing Hermes Agent OS, a system for building your own agent operating system. It includes agents for music, video, SEO, and more, using tools like Miniaax, HM, 11 Labs, and hyperframes.
- **Connecting Agent OS to a VPS** [01:30] — Khaled asked how to connect Agent OS to a VPS. The answer: use a tutorial from John that deploys with Hostinger, secures with Cloudflare tunnel, and gives Cloudflare access. Alternatives include Tailscale or hosting locally for security.
- **Merging Agent OS with Existing Dashboards** [03:00] — Will asked how to plug Agent OS into an existing dashboard. The recommended approach is to merge features one at a time, explaining to the original agent what to bring in from the new version, rather than merging everything at once.
- **Local AI Image and Video Generation** [04:30] — Daniel built a local image generator using Ernie (a Chinese model from Baidu) and a local video generator with AI avatars. It took 2 minutes to generate a video on Comfy UI with an RTX 5090 (32GB VRAM), no API involved.
- **SEO Automation with Agent OS** [06:00] — An SEO setup inside Agent OS can plug in a keyword and case study, then deploy to five websites with unique content. Results show growth from zero clicks to 215 clicks per day for one site, and 72 clicks per day for another.
- **Marketing SaaS on Reddit** [08:00] — The presenter shares a Reddit marketing strategy: post daily to a subreddit with case studies and articles. Views grew from 25,000 to 293,000 per month over 12 months. Content can be generated with the SEO pipeline and posted to Reddit.
- **Using Claude with N8N** [10:00] — To trigger an N8N workflow from Claude, set up N8N as a connector inside Claude's settings. Claude can guide you through the setup step by step, or you can use the connector to build workflows directly.
- **Local AI Engine: Offline Agent Engine** [12:00] — The offline agent engine runs a local AI model (e.g., GPT OSS 20B) on your Mac. It's free, private, and works offline. You can voice-control it to build apps, preview them live, and save to workspace. The model stays warm in memory for fast response.
- **Setting Up Local Models with Ollama** [15:00] — To set up local models: install Ollama (free), pull a model like GPT OSS 20B or Llama 3.1 8B, and keep it warm by setting it to stay in memory for 30 minutes. This avoids reloading delays.
- **Local Hermes Agent Engine** [18:00] — The local Hermes agent engine uses a team of AI agents powered by local models. You give a goal (voice or text), Hermes breaks it into steps, runs tools, and builds the project. It checks if files were actually created before reporting success.
- **Orchestrating Teams of Local Agents** [22:00] — A Kanban board allows teams of local agents to work together. For example, building an SEO blog about OpenClaw. Agents plan, build, and verify steps. The workspace shows all built projects, and you can give feedback or assign new tasks.
- **Benefits of Local vs Cloud Agents** [25:00] — Local agents are private, free, and work offline. Cloud agents cost per token, send data to servers, and can be rate-limited. Local models like Llama 3.1 8B (5GB) are fast and reliable for lightweight tasks.
- **Honest Limits of Local Models** [28:00] — Local models are great for quick code and commands, but not for super complex builds. Use frontier models for heavy jobs and local models for smaller tasks. The engine checks if builds actually land and retries if not.

### Conclusion

The video provides a comprehensive guide to building and using a local AI agent operating system, emphasizing privacy, cost savings, and offline capability. It showcases real examples of automation, from SEO to video generation, and encourages viewers to join the community for step-by-step support.

## Transcript

Captain. [ __ ] >> Good afternoon, sir. Nine open items and nine notes in motion, sir. First up, Vault LLM wiki restructure finalized atomic note layer. In the news, 60% of US consumers say AI in brand messaging is a turnoff. Yeah, that's the drop. Take the Lord. go. Get it. Let's me. All right, we're going to kick this off. Hope you like the music agents playlist right there. MC Goldie on the mic. So, we've been getting loads of questions about Hermes Hermes Agent OS, how to use it. Let's kick this off. So, today we're going to be answering the latest questions on Hermes Agent OS. If you don't know what this basically you can build your own agent operating system and some of the questions inside the community around the one that I've built today we're going to answer because I know if everyone has questions like those then we can answer them publicly and help as many people as we can. And so, for example, you can see recently inside the system, I mean, we've built out all sorts of stuff like a a music agent. We've built out a video agent, SEO agent. We've even got the video agent to the point now where we can plug in a topic, it will go do the research, it will find relevant research to insert, and then it will edit it all together beautifully as you can see right here. here and actually uses a combination of of Miniaax, HM for the avatar, 11 Labs for the voice, and hyperframes to put it all together. It's it's absolutely wild. So, today we're going to cover some of the best stuff that you can do with this. And this is just to inspire you and also just to help you if you're building your own agent operating systems. And if you're not building your own agent operating systems, definitely recommend doing that because it's how you unlock the full power of Hermes itself and any other AI agent that we use. So, let's get straight into this and see what we've got right here. So, first question that we got from Khaled was I have a V Hermes agent VPS from a hosting provider. How can I connect to agent OS? Right. So, we give the zip file for agent OS inside the profit boardroom. How do you connect to it for a VBS? So, we've actually got a a really good tutorial from John who talks through exactly how to use it. So, he deployed it with hostinger, secured it with Cloudflare tunnel and then gave it Cloudflare access, right? And you can see the example mission control that used based on ours uh to create this agentic OS. And so, if you're trying to build your own, give it VPS access, you can do that. You can also use tail scale as well, which is another decent alternative. For me personally, I just host it locally because I think like that's a more secure way to do it. But if you do want to have a system where you can have your AI agents working, you know, from a VPS or working remotely from an agent operating system, this is how you can do it. So, good question from Khaled right there. Next question. So Will was asking how do you plug the agent operating system into an existing dashboard? So let's say for example you got an existing dashboard like this and you're trying to use Hermes agent operating system to plug it in. How would you connect them etc. So the way that I would approach this obviously they're two different systems and you're trying to merge them which is a little bit complicated but the way that you would do it is you explain to the agent whichever agent built out your dashboard originally. Hey here's a new version of the dashboard. Here's an old version. I just want to bring in some of the versions and some of the features from the new version inside the Aentic OS into my old dashboard and then just slowly build out like one feature at a time. I wouldn't merge a whole thing at once simply because if you do the whole thing at once, it's going to be messy. It's going to be complicated. Whereas if you simplify it down to like just a few things or or one thing a day for example or one thing a week then you can merge it slowly and you can build our agentic operating system into yours. The other option as well is of course like you could have sort of you as we were talking about before with a VPS gateway you could kind of have that linked to the client dashboard as well. That's another option but I think that would be even more complicated. So I'd stick with version one. By the way, if you want to be able to ask me questions and also get our agent operating system, you can get that inside the AI profit boardroom link in the comments description or just go to the apiprofitborn.com. Daniel had a pretty cool result with a local AI setup. So basically he's been building out a image generator that works locally which is pretty wild when you think about it using Ernie which is a Chinese model from Badu. So it can generate images locally. So if you're looking for like a local image generator you can actually get this off hugging face and then generate images locally which is wild when you think about it. So, one question we got here was about how to trigger an NASA workflow from Claude. So, let's say for example, you've got NA10, you've got Claude, you can actually set up NA10 as a connector inside Claude. and they have full example of how to do it as you can see right here. So that's probably the easiest way. Then you can just go to your settings connectors and add N8N as a connector inside your settings. And if you're struggling setting it up, you can just ask caller to guide you through it step by step. Now this is wild. So Daniel has his own local setup and basically what he's done now is he can create videos locally with AI avatars and you can see an example right here. So he's talked about setup and how to do it and bear in mind like there's no API involved with that, right? It's just a local setup which I mean if you can generate images, you can generate videos locally etc. This is absolutely amazing. And it took around 2 minutes to generate this specific example. Let's have a look at this. Wow, that's really good quality as well, isn't it? This was on Comfy UI with an RTX5090 with 32 GB of VRAM. So that shows you what's possible with this stuff. So Claire has just joined and she was looking for help with SEO and how to automate a websites with SEO. So one of the things that we've been looking at recently is this SEO setup inside our content pipeline. It's inside the Aentic OS. And basically what you can do is you can plug in a keyword, you can plug in a case study, and then just deploy to five websites with five unique pieces of content. And these are the example websites we've been deploying content to. So you can see an example right here of what it looks like and how it works. And if you're wondering if it actually works or not, you can see an example of our results here as well. So this website, for example, it went from like one or two clicks a day. Let's have a look. Is that zero clicks a day over there and then it grew to 100 well 215 clicks a day recently. It's also on a super nice trajectory. Here's another example. So, this website grew from like zero clicks a day all the way up to it's getting regularly like 72 clicks a day and it's growing in trajectory as well. So, you know, this process basically what we've got here is a SEO skill that can create content step by step using our system here. And it's super nice, great quality, a nice little skill and it's plugged into the Aentic OS. So, that's been pretty amazing for SEO. Definitely recommend it. The cool thing is well you can see here for example like Bobby building out his own custom GPT for the first time. So there's all sorts of people at different levels like you got some people running their own super complex local setups that can generate AI videos and images and then you got other people who are just joining but it's the first time they're trying AI and they're still building like pretty amazing stuff. So, Now one question we got from uh RPA was like how do you you know how do you market a SAS right? So especially on Reddit. So this is what we do cuz we do this for the agent OS system and also the air offer boarding. And basically the way this works is we have a subreddit which we post to daily and you can see like per month. Look at the last 12 months it's grown from like 25,000 views a month to last month we got 293,000 views in the last month. And so what we do is we we post content daily as a team to this subreddit and basically post articles with case studies. So you can see the case study right here and then you can see the full article underneath and that's a really good way to do it. You know you can reach a lot of people on Reddit and you can automate this with AI as well. So really powerful system right there. And you can also generate the content with the SEO pipeline, but you would just, you know, post that to Reddit instead of your websites directly. Also, one of the other things about this as well is like you can actually update and build new things in into your agent operating system whenever you want. So, for me personally, I build these workflows in just because I'm like, well, okay, this would be super useful. you know, the video agent, the SEO agent, music agent, uh, game studio even. And so, anything that you find useful, you can build in here. So, if you look, for example, uh, Dan, he looked at the open router fusion situation and then he plugged in his system that could do all of this stuff as you can see, and we've already built fusion in here as well, which is kind of like a way of achieving Fable 5 level intelligence according to the benchmarks. I would always try it yourself. But according to the benchmarks, you can achieve Fable 5 level intelligence with smaller models. So if you not tested this out, it's a good it's a good thing to try and you can always build new workflows and just adapt it however you want. So if we look at the screenshot here, you can see how they've adapted their mission control to build in this automation. So anything that you have an idea for, you can just build it in with your AI agents. So this is another question similar to the one before which is like can you build the same sort of automations with chord that NAM performs and the answer is definitely you can. So yeah, you can definitely. So for example, the way that you can do this is you've got two options. Number one is you can go directly to Claude and Claude can build any of this stuff out for you directly. It would just guide you through the process if you explained exactly what you want. Option number two is you can actually use NA10's connector with Claude and control NA10 and build workflows with Claude directly through the connector. So there's two options for that. Now, here's a question from Brian, which is, you know, should you get into Hermes or not, right? What's better? The way that I would look at this is if you if you don't have the understanding of what the difference is between these agents and that sort of thing, and you're if you're happy with codeex, I wouldn't switch because you can just stay 100% focused and there's no need to switch. Like it's fun to use Hermes agent, but if you don't have that much time and you want to avoid overwhelm, I would just keep it simple and stick to codeex. So we got a question from Lloyd on how to create like AI avatar videos. So the way that we've been doing this recently is using that video agent as you can see here inside the agent OS. So you can see an example of how that works and that is a fully you know AI powered video. But we've also got full training on our process which you can see over here. So, if you want like a step-by-step breakdown without using the agent OS, you can use that, too. So, there's two different options for that. Now we got some SEO questions here from Paul. So Paul was saying, you know, recently launched a a travel website, low domain ratings, though. If you don't know what domain rating is, it's just the authority of a website. And three questions here. Are higher quality landing pages as important as blog content? So for number one, I would say it's always coming down to the keywords you're trying to rank for. So for example, if you're trying to rank a landing page, then the quality of that landing page rank matters a lot. The same with blog posts. If you're trying to rank a blog post for a potential keyword, then the quality of that blog post matters just as much. And when it comes to like keyword cannibalization, you don't want to create two types of content for the same keyword. So what you would do is just look at what's already ranking, reverse engineer that, and then create content around that. Usually the AI can actually figure out the search intent itself. Now, for AI search visibility, is LLM's text worth implementing? For us personally, we've not implemented LLM's text. It's not that important, and we're ranking pretty well inside AI search engines. Here's an example. So, you can see for this keyword, we are ranking here, right? and we're also ranking there too and also there. So you don't need LM's text to rank for those keywords inside AI search engines. David was talking about GLM GLM 5.2. It's available and it's US hosted on Alamas Cloud. So, we've already built it in to the agent operating system as you can see right here. And it's built some like absolutely amazing stuff. So, for example, you can see an example what we built here. If we keep going through, here's some more examples of stuff we've built just for fun, but just also just see the power of it. And it is absolutely amazing as a tool. It's not as good as Claude Opus 4.8, but wow, it is close. Very, very close. So, super powerful stuff right there. And really good to build with. You can also use it inside uh inside um Hermes as well. You can plug the coding plan into your Hermes agents. Another question we got was about paperclip. Now paperclipip is a system where you can basically create like a team of AI agents working in their own sort of AI company like you can see right here and you can build some amazing stuff with this. You can see all the stuff we built here. Now, if you're struggling to set it up, paperclip is a little bit buggy. So, we actually got a full fix right there inside the air boardroom on exactly how to fix that, which is great. And I think that's all the questions answered on the agent OS. So that is Hermes agent OS. We've talked about some of the best ways to use it, local setups, how to generate video and images locally, some ways to use NAN with Claude, some of the best automations as well, and how to integrate this inside one single place as you can see right here with all of our different workflows. Now, if you want the full setup from me, you can get that inside the AI profit border link in the comments description or go to a profitboard.com. And inside the prof you can ask me questions like this and I answer them in video tutorials like you can see every day today. So I write down the whole process step by step and then I also answer them inside the videos you've seen inside the community. You can ask questions, get help and support whenever you want and every question gets answered which is great. We've also got 106 people online right now. So 24/7 we've got people online ready to help you as well which is great. It's a great community with all of our best trainings and you can get new daily updates over here. So, we had new daily updates based on what's just dropped and we have the full agent operating system which we literally just updated today and you can grab the full zip file right there. You can also jump on weekly coaching calls where we wire it together and build it together and then inside the map you can meet people in your local area who are using AI agents like you. So, feel free to get this all link in the comments description or go to the aiprofit.com. Thanks for watching. Let's see what we got here. I think some AI models want to lead us round of circles. Yeah, is it happens, right? So, one of the things that we built in to fix that is we have a loop system with Hermes agent. I've got a tutorial coming out on it today. And basically what you can do here is you can describe what your definition of done is so that your agents don't loop around until you know nothing gets done. And then also have one model that runs on free APIs here and you can set how many rounds it goes for so it doesn't loop around forever. And then we have a judge that looks at the work, judges if it's actually good or not and then sends it back if it's not good enough and they just iterate round and round. So, if you've not set up that system yet, definitely look into this. Again, we've already built it into the agent operating system, but yeah, it's pretty powerful. I'll be dope, mate. I'll be dope. So today I'm going to show you how you can basically build and automate anything with local models. And this is something I've tested. I've tried a lot of different local models recently. I'm on a Mac Studio, but you can use whatever you want. There are some models that are great. There are some models that will slow everything down. I'm going to talk about the the benefits of using a local model, how to run them for free, how to run them privately, how it works, etc. And what we've built as a system to basically automate anything. Now, as an example of this, this is our local AI engine, as you can see. So we can actually voice control this and we could just say something like okay build me a to-do list app and then we can control it with our voice. We can click send here and what it will do is start building and we can actually preview that once it's done right which is pretty powerful in itself. Now also if you want to see okay what have we built in the past with this stuff. So whilst that's thinking and building locally I'll show you some stuff that we've actually built with this. You can see here for example all this stuff that we built. Um we can open this up and this was just like for fun but you can see how you can actually build things and then save it to your workspace inside this agent operating system that we built which is quite a lot of fun. And also the cool thing about this as well is that once we built it so you know we just asked it to build a to-do list a section ago. You can see that it's actually ready to go and ready to preview here. Now, this is not like you're not going to get like frontier level stuff unless you are unless you have an amazing setup. And I'll come on to that in a second. But you can see how this to-do list actually works. We can control our voice. We can open it full screen. It's going to be saved inside our workspace. So, we can preview it here as you can see. That was literally just one test. So, you've seen a proven case study of how this works in real time. You've seen how we can operate with our voice. You've seen how we preview it and also workspace. Now, the other cool thing about this well is like if you have a really good setup, let me show you an example. So this is Daniel inside the profitable boardroom and he actually built out a full avatar video which is crazy like a full AI powered video with no APIs because he built it locally right on his setup. Now this is using uh RTX 5090 with 32 GB of VRAM but that just gives you an example of the power of this stuff if you have the right setup. So let me talk through exactly how it works. I mean you can see another example here. So this was a image generator from Daniel and he's basically been building images locally using AI and local models, right? The same local setup that we just talked about a second ago. So this can be super powerful if you have the right system in place and that's what we're going to talk about today. Uh, so this is the offline agent engine, which means that you have a real AI model running 100% on your Mac. It's free. It builds instantly as you saw a second ago with the to-do list app. And also, it's completely private. So, normally, for example, if you're using an API, it's in the cloud, which means someone else's server. Also, usually you have to pay for API and it's also your data going to the cloud. Whereas, for example, if you use a local model, you can say, okay, me this, it loops back. is on your Mac or wherever you're using it and then you get a quick answer as we saw a second ago and nothing leaves your Mac. It's all local and private. Like if you don't have Wi-Fi, well, no problem. You can still use your local setup. Now, one thing that I found is that it's a little bit slow to set up a local system. Unless you have it running, which is what we actually set up with this system here. So, we can just build stuff privately, locally for free. We can do it offline as well. and we can save everything inside the workspace. And because the Asian OS is an offline system, you know, it's hosted locally, it doesn't matter uh what happens, we've still got it ready to go. So, you have the system and you can just swap models in and out. Now, if you look at the old way versus the new way, right? A lot of people will have a subscription. They've got a a monthly AR bill. Uh, every message you send leaves your computer. Sometimes you hit rate limits. You try a local model, but it lags for seconds, especially if you're setting up a new one. You blame the model, give up, go back to the cloud models, and the result is that you end up using an AI that's not local, that's not free, that's not private. The new way is you can use the offline agent engine and you pin the model warm once so that it never needs to reload again because it's inside your agentic operating system and it's ready to go whenever you need it. It also answers pretty quickly as we saw it built that app very quickly. You it's completely free and you can run it all day. You know, it could be building for you 24/7 if you wanted and it stays private so nothing ever leaves your computer. You say the word and it could build the whole app. literally you can talk to it and the result is fast private AI that you actually own which is absolutely wild when you think about it. So you also might be thinking okay sometimes local models are slow and it's usually the loading. So if I open up for example OAMA and I try and get Gemma 4 to run right now even on a Mac Studio it's going to slow everything down because it's got to reload every time. Whereas for example if you have this set up pinned warm the lag is gone. So, what we actually did is we have agent uh GPT OSS 20B running here. And that out of everything I've tested seems to be the fastest and the most reliable, which I was pretty surprised by. I thought like Gemma 12b might be good on a Mac Studio. It's it's super slow. It just slows everything down, especially if you're running local models. I think the the interesting thing here is like you don't need Frontier intelligence most of the time. Like if you're building and you're building locally and you want something offline and free, like you don't need the most powerful model, you know, you can always come back to to to one of those later, but it's definitely worth testing this out. So, out of everything that I've tried, GPT OSS 20B, which surprised me works the best. And I've tried genuinely like everything. I've tried uh Quen, we tried, for example, Gemma 4, and there was a bunch of Quen models that we tried, but this one seems to be the most reliable. Now, we can run it offline, which is pretty easy, and that's easy to say, but basically, you can have a prompt running on your GPU, and then it answers back to you, right? And it's sealed because it's all private. It's not going to the cloud, it's not going to the internet or anything like that. So, your prompt, your model, the answer, they all live inside this box right here, which is pretty powerful. And you can see an example of the test here. We actually checked like where does it go? And we tested it and it was all good. Now, the other cool thing about this is you can say what you want and watch it appear. And this kind of feels pretty futuristic. You don't even have to type at this point. You can just, for example, go into this system here and we can preview anything that we built previously. And then if we click on the voice over here, build me a neon countdown timer that glows. just as a quick example. And if we hit send now, it's going to start running as you can see, right? So that's going to start building locally for free, which is pretty amazing. So you can speak it. It writes the app or the website or the tool or whatever you're trying to build. You can preview it live because it's going to come up with a preview once the HTML is fully written. And then it's saved in your workspace. So you can open it up in any time. And there's three tabs inside our system and we've got the countdown here as you can see just as an example and it's actually working which is great. So we've got the build tab where we can talk to it. We can type what we want and it's going to build directly. We can see what we've built and preview it and we've got the history the conversation history. So if we refresh a page like so you can see that the conversation history is saved which is super useful. Then it's got the preview over here. And then it has the workspace with everything we've just built, which is great. So it actually works. And honestly, like when you see a lot of people using local AI, they don't actually show like the outputs or what works. This is actually good and it's actually creating interesting stuff, which is great. Now, if you want your own offline agent engine, I've actually wired this into the agent operating system. The dashboard where all my agents live on one screen. The local model is just one more agent in there sitting right next to the cloud one. So, we've got the full agent operating system with every agent in one place. The UI that you just saw, the dashboard, etc. A 30-day road map to wire this in fast. Free with private agents actually do the work. Four coaching calls a week with people already running this as you've seen from our members and a room of 3,600 builders. So, there's always someone online ready to help you. So, you can get that in the link in the comments description or go to the aiprofitborn.com. So let's talk about how long it takes to do this yourself. So literally all you want to do is pick a model that fits your max memory. So for example, for us on the Mac Studio, GPT OSS 20B seems to be the sweet spot. We tried a lot of different stuff. We wanted something fast that was a mixture of experts that sits ready to go at any time. You could even go even lighter than that. So, for example, Llama 3.1 8B works pretty well as well. But you obviously you can't run like a giant model on Mac Studio because it will just slow everything down. So, for example, if you got like a 28 to 30 GB model on a 36 GB Mac, bit of an issue. Slows everything down. Doesn't run smoothly. And so, to do this, what you need to do is step number one, install O Lama. It's a free app that runs models on your machine. You can get it from.com. So you can get that free over here. You can just download it and then you pull in the model once. So this will download it directly and then once it's done it's ready to go. Now you can see inside O Lama there are cloud models there are offline models. You want to go with an offline model. So for example if we go to this one this is not a cloud model. We can download it and run it. Pretty simple. And then what we've done is we just keep it warm, which means the one setting tells the model to stay in memory for half an hour after you use it. So it's just warm and ready to go whenever I'm using the agent OS. And I I just tend to find that works better because we don't have to reload it every time. And then we can wake it once. So if we send it a quick hello, it will load into the memory and then you're good to go on this example of how to do this with Lama as you can see right here. And that's the whole engine. Now, if you've got less memory or you've got a less powerful setup, well, you could just run a smaller model like Llama 3.1 8B that's only 5 GB, 64K context. Runs the same way. You can just run it for like lightweight tasks, you know, or you could run it inside Hermes agent, for example, and it's just good for the quick stuff. Now, some people are going to say local models are too slow to be useful. Honestly, I believe that up to this morning when I tested it out and we tested many models, a lot of them were really bad, really slow on a Mac Studio, this was the only good one that we found. So, they're not slow. Usually, they're quite cold as well, or they're just taking up too much RAM. So, it depends on your setup. Uh, other people say, well, a free local model is not good enough to do real work. But, I've shown you proof that actually can build interesting stuff. And then of people say, well, setting up a local model is too technical for me, but it's just one app store with three short commands inside O Lama. So if you can copy and paste a couple of lines, you can run this. The hard part is actually making it fast. And again, I've shown you already like member wins from the AI profit boardroom who have done this, but you can see that we've actually got over 183 pages of testimonials from people building with AI agents, building their own agent operating systems like I've shown you. So, if we can do it, you can do it, too. That's what I want to say. And that's basically the whole setup. So, if you want to get a full local AI, you've seen it. you. It runs online offline. It's free. It answers very quickly. You can build whole apps when you say a word. If you want it set up with you step by step, right next to your cloud agents is all inside the AR profitable boardroom. So you get the full agent operating system, local models, claw, gm, Hermes, and more. All inside one dashboard with shared memory. You get the offline agent engine setup. So the model um we've configured the settings for like the warm pin and the voice build surface as well. You get a 30-day road map, daily tutorials, and four coaching calls a week, and a room of 3,600 builders doing this as well. So, if you want to join me and everyone else inside here, you can get it. Link in the comments in description or go to the arr.com. Inside the community, I personally answer questions. So, you can see, for example, every single one of these. Number one, everyone else in the community helps each other, but number two, we I post personally as well. So, this is something that I get involved in and I help everyone inside the community as well. inside the classroom. You can get access to all of my new trainings so you can get help step by step and and get everything set up. You can also go from beginner to expert with this setup. And also, if you want to see the agent OS system, we've got a video tutorial on how to set up the last update. As you can see, we got a full guide on how to use it and then also the resources at the end as well. And we update that daily. So, we had new improvements every day. which is built in the local engine for using local models which is super useful. Plus we add new tutorials with video tutorials. Like you can see inside the calendar you can jump on weekly coaching calls. We have four coaching calls a week where you can get help and support in real time. And inside the map you can meet people in your local area who are using AI agents like you. So feel free to get that link in the comments description or go to the aiprofitborn.com. Thanks for watching. Good to see you here, Drew. Thanks for joining. Did you come up with a system to bring all your different LLMs together to work on projects together? Can they speak to each other in real time? Yeah. So, there's four options for that. This is the easiest one. So we can for example go inside our AI agent group chat and we can ask our agents to talk together. So for example here we can say okay give me a great idea guys for an SEO app. So we type that in and then Claude is going to come back to us and we basically get an answer from our whole team. So if you want your agents to talk to each other you can see an example of how that works directly here. So, we're now bouncing ideas off each other. As you can see, what we've also got is a pipeline. So, our agents work together to plan, build, I just approve it or not. And then you can see everything that I've built over here. So, that's another way. So, there's two options. Number one, it's a group chat. And you can see your history here if you want to see stuff that's been done already. So, for example, Codeex added that SEO app idea to our pipeline. And then we can build it over here. So, it's like a agent to-do list. And then we also have paperclipip and we have a local agent camb board where basically our offline team can build together. Can you connect to notebook? Yes. So we have this section over here where we can basically automate whatever we want with notebook. It's already connected via MCP from my agent operating system to notebook. So for example, you can see my full library over here. We have the research section where if we plug in a tool like this, we can do research on it very quickly. And then also we can chat with it. And we have a studio where we can generate audio overviews or videos or slide decks or mind maps or infographics, flashcards, quizzes, data tables or reports with this. And then finally inside the asset section, you can see for example all the stuff that we've built out and pulled in. So for example, we've got like videos, podcasts, infographics, all automated with no glm plugged directly into our agent operating system. So can you orchestrate teams with it of AI agents? Absolutely. As you can see, can you link it to notebook? Absolutely. As you can see, right? Um I've really built it based on like all the feedback that I get from members inside the air profitable basically on what they ask for. I build it in to help you as much as I can. So yeah, it's all inside here. Well, do you have any videos on how to build a system like this? Yeah. So, you can get it inside the arprofit boardroom. So, if you go to the arrroom.com, if you go to the classroom, you can actually get the exact system that we have inside here. So if you go to the agent OS system, you can see the video tutorial, the last update date, and then the zip file to install it. So if you want to get our full system from us, you can get it inside here. Today I'm going to show you how to build a local powered Hermes engine which is a team of AI agents powered by local free private AI and you can basically automate whatever you want with it. So this is something I call the local Hermes agent engine. You can give it a goal that could be a voice or text. Hermes breaks it down into goals and steps. Then it runs the tools with commands and creates the project for you and it's built and verified. Plus you can come back to it inside your workspace, right? And the great thing about this is number one, it's private so your data is not going to the cloud. Number two, it's free because you can and you can have these agents working as a team building stuff 24/7. And number three, it is completely running locally, which means, you know, you don't need the Wi-Fi or anything like that to run it. So you can give it a go. It plans, it runs the tools, it builds the thing, every step on your own machine. Pretty amazing. How does this work? So you can see an example right here. This is the agent camb board, local agents. This is a team of local offline agents with Hermes working on a live camb board, separate camp board just for them. And so I could say for example build out an SEO blog about openclaw which we've typed in here and then it assembles the board. So what it actually does is it comes up with the ideas and what it needs to build for this website. Now if we click on for example run the team what that's going to do is start building this step by step to create whatever we want which is pretty amazing in itself. Now if you're wondering okay like what does this actually look like in action? Well you can see them building now. So these are AI agents running locally with Hermes to build and automate whatever we want. And then if we go over to the workspace here, we can see the stuff we built previously. Um we can also see the prompt. We can open this up full screen. You can see an example. These are just simple examples just to show you what's possible. I literally built this this morning into our agent operating system. And you can see it's now creating the landing page and everything else using our AI agents together. Now we can also open this up. We can give it feedback. We can tell it to refresh it. We can ask it to improve everything that's built. But that's basically running as a team as a loop together, which is pretty amazing. So you can see some examples of what we've built with them. And each of these just comes from a single sentence. So this was built by a model running entirely on a Mac. No internet, no API costs, just Hermes agents, teams of agents working with local models. Now, if you're running, for example, OAMA with Hermes, you can get this set up pretty quickly. So, there's a couple of options here. Number one is you can actually go to Alama. Just make sure you have OAMA running in the background. And then from here, you would go inside whichever model you want to run locally, for example, like this one. And you can just copy and paste this command to run this inside your terminal with Hermes. However, if you want teams of agents, that's why we've created this camb board here. So, we can have a team of agents working together to build and automate whatever we want, right? And so, the great thing about this as well is like we can speak to our agents inside this section as well, which is great. So, we've got like for every API, we can change the model and we can create a separate agent profile. We can also talk to Hermes. Here we have Hermes Javis which is a voice activated version of Hermes. We have a studio where we can generate images, video and voice. Bear in mind you can generate images from local models like Ernie from BU. Then you got session section and the workspace here with everything you've built with each model broken down as you can see. So we can see everything that we've built previously with Hermes agent as well. Now if you go to the manage section here and we go to models, you can also change the model this way. So for example, you can see that we've got LM Studio ready to go, which we could run local models with. And then we can also run it with Nvidia. Nvidia is another way to run local models. And of course, Olama. OAM is another way to run local models. So we've got OAM ready to go with two different models over here. As you can see, you can run Gemma 4 with this for example, and you can have teams of AI agents working together. Now, if you want to visualize the teams working together, you could have a camb board here. So, if we go back to this system, you can see everything is built now, which is pretty nice. And we can just see everything that we've built inside the workspace here. But we can also assign new tasks and get them working together again on new tasks and just keep building these teams out, right? So, picture this. You can have a team running 24/7 building new useful stuff on a camb board. You can see exactly what's built. You can view it inside the workspace. And this is all free, private, and offline. Crazy stuff when you think about it. So, how does this work step by step? Well, if you look at this whole system, it's just one goal in and a finished job out. So, you don't hand it a checklist, you hand it the outcome. So, you could say, okay, list the files in this order and write me a summary. It doesn't answer in words. It runs a command, reads a result, writes the files, and then tells you once it's done. And that loop, plan a step, run a tool, look at what happened, plan the next is what makes it an agent instead of a chatbot. So you give it the goal, it runs multiple agents. So for example, it will run one for running, one for reading, one for writing, and then it's done. So it plans the steps, runs each tool, checks the result, and only says done when the file was really there. So you can see an example of a real run that we did with this system. Now you might say why do this? So if you look at this particular if you're using agents like agents they use up a lot of tokens. So a lot of people want to run stuff and the thing is if you're not running local models then every command and file is sent to someone else's server. You pay per token on every single run. If you have no internet then you don't have an agent. It's just going to stop. You get rate limited sometimes and then you get cut off mid job. We've all seen that with Claude. you trust it, built the thing, but you can't see it. And the result is a powerful agent that you don't really control, you know, and we've seen that with Fable 5. Like when Fable 5 got taken down, loads of people's workflows just ended. So with the new way, every step runs on your Mac. Nothing is sent out. It's free to run all day. It can work on a plane, in a cafe, with the Wi-Fi off. It breaks a goal down into steps as you saw and then does them itself. It checks a disc, shows you exactly what it built, and the result is a real agent you own, free and private, that can run 24/7. So, the work is the same, but the difference is where it runs, how much it costs, and who sees your files. That's a big difference between the old way and the new way. Also something to be careful of is like local models. Sometimes they say they built the thing but they've not actually built the thing. Like it's very common if you've ever run a local model. So with this system what it actually does is the agent says I built it and then it checks did it really build a file. If it didn't nothing lands inside the workspace. If it does then you actually get everything and it and it builds um as you can see. So everything here was built with local agents. Now if you want this local Hermes engine with the agent operating system, it lives inside our agent OS. So the dashboard where all my agents sit on one screen. The local agent runs right next to all our cloud ones, free and offline, ready for the quick stuff. So you stop using up tokens. A full agent operating system with local and cloud agents plugged into one dashboard. with builds previewing live. You also get the local Hermes engine set up, the profile, the model, the workspace, etc. A 30-day road map to wire agents to do real work for you and a room of 3,600 builders, right? So, there's always someone online ready to help you. So, you can actually get that inside the AI profit boardroom. Link in the comments description or go to the profitborn.com. And if you go to the classroom, you can grab the agent OS inside this section. As you can see, if you're already a member, by the way, and you're watching this, you're like, "Well, how do I add these new updates?" So, we have inside the new zip file, which we update daily, we have an update section where you can add the new updates and import them into your existing setup. So, if you want to update this, you're already a member, no problem. You can do that using the setup and the zip file that we give you inside there. So, this is basically it. Now, let's talk about wiring it in, what models we've used, etc. So there's two pieces to this. Uh local model that fits your Mac and the Hermes profile pointed at it. Hermes need a model with a big enough context window. So you could use for example like even I mean we we as this example we've used Llama 3.1. You could also use GPTO OSS. I've tested a lot of local models and unless you have a really good setup many of them are bad and super slow. So these are the two models. GPT OSS Llama 3.18B that have worked for us on a Mac Studio. And it really depends what setup you've got. Like for example, this uh someone who was posting inside the profit board earlier, Daniel, who's an absolute legend, and you can see an example of what he generated using the local model as well. That's with an RTX 5090. So, it's pretty amazing what you can do if you have a really good setup. If you don't have an amazing setup, you can just use lightweight models that I've shown you today. So, you can wire it in in about 10 minutes. You just pull in a light capable model. So, for example, like Llama 3.1. It's only 5 GB, runs on most Macs. It's fast and crucially for an agent. It can use tools instead of faking using the tool. Right? A lot of local models say they've used the tool, but they've not. Then you make a Hermes profile, point at it, so you have a dedicated local profile the offline agent uh uses and you can point it at the local model and keep its context window. You can keep it ready to go. So quite often if you're swapping or if you're opening up local models and then shut them down, it slows everything down where with with this system you can have like Llama 3.1 ready to go at any time and then you just give it a goal from the KBA board which you can see right here. And it's a really powerful way to orchestrate your agents locally. I mean, we have a system inside here where we can have a group chat between our agents and they can be orchestrated like this. We have the pipeline where we can automate and build with teams of agents anything that we want. And we've got a full gallery of what we've created here. It's like a AI agent powered to-do list. But if you want local agents building together, this is the best way that I've seen to orchestrate local agents running together. You can see an example of the four steps right here and how it works. Now let's talk about the honest limits here. This is great for running commands, firework, quick code, etc. Not great or super giant complex builds. So a small local local model is like a fast helper. It's not flagship or frontier model. You can hand the heavy jobs to a frontier model, but then the smaller jobs you can give to your sub agents or to Hermes agents or whatever and it checks itself. So when a build doesn't land, the engine tells you and then it just runs again. So some people say, well, a local model can't do real work. You've seen examples of actually building stuff today. Other people say, well, if it's free and local, um, it's not actually going to be useful. Again, we can actually get them running with a team of agents using this agent cambam, which is pretty amazing. Then other people say, well, running agent locally isn't that too slow. And that's only the case if you run a model too big for it. So you want to pick one that fits. And I've given you a couple of options today. Now, if you're thinking, "Okay, like that's great for you, Julian, but what about everyone else?" You can see that we have over 183 wins and uh pages of testimonials and wins and reviews right here from AI Profit Board and members. So, I know that if I'm not a coder and I can get results with this and you're not a coder, then and these members are not coders, then we can all get results with this stuff, right? We're at the point now where you can just talk to AI like a friend and it builds whenever you want. So, that's basically the whole system. Now if you want to get this agent operating system from me with paperclipip builtin the AI agent mastermind the pipeline the agent camb for local agents claude openclaw hermes gemini codeex kim code gm 5.2 to gro build free claw code. We got fusion inside there. Uh local agents. We got a self iterating feedback loop for loop engineering with Hermes agent. The SEO system for deploying content. A video agent for just building and creating amazing edited videos like you see with AI. It's got my avatar inside there too. We've even got a music agent, a game studio, uh notebook builtin, camb boards, and a full memory system. That's all inside the AI profitable boardroom, which is my community focused on helping you scale and save time with AR automation. And this is a amazing community of 3,600 members. And you know, if you ask uh if you ask a question inside the community, we get back to you daily. So, you can see, for example, like Brian asked a question seven hours ago. We've already replied to him, right? And helped him. And he says, "Got it. Thanks." Right. So, if you ask a question, we get back to you pretty quickly. I make video tutorials helping members inside there daily as well. You get access to all of my best trainers inside the community. And inside the new daily update section, you get access to my agent OS system with the video tutorial, the last update date, the zip file completed so that you can install it. New daily tutorials like you can see a calendar with four weekly coaching calls. So, you can jump on this coaching calls, ask questions in real time, get help and support whenever you need it. inside the map. You can meet people in your local area who are building with AI agents. And this is all available link in the comments description or just go to the aiprofitborn.com. Hope to see you inside there. Cheers. Bye. Thanks. If you want to connect with me, uh the only place I really answer stuff is inside the the community, the air profit warning. Yes, you can watch it later if you want to as well. All right, I think that's everything for me. Thanks for watching. Cheers. Bye-bye.