---
title: 'Free Local AI Model, SEO Agent OS, and Token Saving Tips'
source: 'https://youtube.com/watch?v=0ppnSyQHUYw'
video_id: '0ppnSyQHUYw'
date: 2026-07-14
duration_sec: 0
---

# Free Local AI Model, SEO Agent OS, and Token Saving Tips

> Source: [Free Local AI Model, SEO Agent OS, and Token Saving Tips](https://youtube.com/watch?v=0ppnSyQHUYw)

## Summary

This video provides a comprehensive overview of the latest AI tools and techniques for building agent operating systems, including a new free local model called Agent A1, an SEO agent office, and a token-saving method called Caveman. The presenter demonstrates how to integrate these tools into a cohesive system for automating workflows and reducing costs.

### Key Points

- **Introduction to Agent A1** [00:00] — Agent A1 is a new free, local, open-source AI model from Shanghai, designed for agentic tasks with multimodal understanding and tool use. It is a 35 billion parameter mixture-of-experts model that only activates 3 billion at a time, offering fast response times.
- **SEO Agent OS Overview** [01:30] — A new open-source SEO agent operating system called SEO Office allows users to visualize and manage a team of AI agents for tasks like keyword research, technical audits, and content creation. It features a second brain for memory and an orchestrator agent.
- **Caveman Token Saver** [03:00] — Caveman is a free skill that reduces output tokens by up to 69% by instructing AI agents to respond concisely without filler words. It can be installed with a single command and works across multiple agents, saving costs on API usage.
- **Omniroot Free Coding Engine** [05:00] — Omniroot is a free open-source AI gateway that routes requests across 237 providers (90 free) with automatic fallback. It allows coding for free forever by using a single local endpoint and supports multiple routing strategies.
- **Free APIs and Models** [07:00] — A GitHub repository lists free APIs and models from providers like OpenRouter, Google AI Studio, and Nvidia. Users can plug these into their agent OS for cost-effective AI operations.
- **Agent OS Features** [09:00] — The agent OS includes Hermes agents (Oracle, Astros, Apollo), a memory system, and integration with local models. It enables automation of daily tasks, content creation, and coding.
- **Testing Agent A1 Performance** [11:00] — Agent A1 was tested on 42 projects using Goldiebench and scored second among local models. It successfully built games, landing pages, and other applications, outperforming some competitors in speed and quality.
- **SEO Office Workflow** [13:00] — The SEO Office allows users to dispatch up to 25 specialist agents simultaneously, perform technical audits, and generate SEO-optimized content. It integrates with DataForSEO API for keyword research and backlink analysis.
- **Caveman Installation and Usage** [15:00] — Caveman is installed via a one-line command and can be toggled with /caveman. It offers light, full, and ultra modes for varying levels of conciseness. Tests showed 69% fewer output tokens and 37% cost savings.
- **Omniroot Setup and Benefits** [17:00] — Omniroot is set up with a terminal command and can be pointed to Claude Code. It provides one endpoint for all tools, auto-fallback when limits are hit, and token minimization of 15-95%.

### Conclusion

The video demonstrates that with free local models like Agent A1, token-saving techniques like Caveman, and routing tools like Omniroot, users can build powerful agent operating systems at minimal cost. The key is to focus on automating essential tasks and personalizing the system to individual workflows.

## Transcript

Good to more good peeps. Today we are going to be running through a new AI model that is free and local and open source. It is called Agent A1. So I've just been testing out. It's actually slightly impressive. um especially for an a local model seems to be pretty fast to respond, pretty easy to use. Um so I'll be showing the results of that in a second. We're also looking at a new SEO agent OS which looks a lot of fun. It's a new open source project I've been testing out and we'll be running through a few other things as well. So should be good. If you got any questions, feel free to ask. And we're just going to start building out on the agent OS as you can see inside Claude. Good day to you, Jay. Hope you're having a good one. Welcome. Stop. Jamie, welcome here. Thanks for joining. Also going to be testing out something called Caveman. Looking how to save tokens with caveman. explain how it works and test it as well. showcase each test it and then also what we need to do is make sure you explain in a simple way so that anyone can understand this Hey hey hey. go and use it inside free claude code. Just talk about how you can use it inside claude. Can you explain what you mean about a new agentic system? So, this is something that we build out every day. have a system where we can use all our agents together, right? And so this is what we're talking about when we talk about agent operating systems, how we use it and that sort of thing. And then we're just building new features into it every day. So yesterday for example, we created Hermes Astros. We can look at our competitors workflows and give us suggestions based on that. And then we also have, for example, Hermes Oracle. We have Hermes Apollo which is a voice activated version of Hermes. So this is how we use it. Quick question. Do you think Fable 5 best use case? Um I mean like you can use it however you want. The the best way to use Fable 5 the best way to get value out of Fable 5 is look at what are you currently working on and then how can you use how can you automate that with Fable 5, right? So for example, if you do a lot of SEO, create an SEO tool with Fable 5, right? If you do a lot of video, create a video agent with Fable 5. So you want to reverse engineer what are you working on because everyone's use cases are unique and then based on that look at how can you save yourself time. Joe, thank you very much for joining, sir. Look at that. Switch to Opus 4.8. Only been using Fable 5 for 2 minutes. It's a classic. You set this up inside the agent OS SEO section. Make sure you configure it properly. uh make sure also that you have configured it and make sure it works. Also create a guide on this screenshots and additionally make sure that you use the data for SEO API that we already have plugged in to open SEO. So you already have the API for that three members. Uh, we have a lot of training on that in the air buffer board. If you want to know how to play like really good AI after videos, check out the AR profer board. Shows you exactly how to do it. This is no bueno. Look at this. Opus 4.8 services busy. It's a daily challenge we've brought. Thank you very much. Appreciate it. I appreciate that. I hope you have a great day. Hey, hey, hey. Heat. Heat. N. Heat. Heat. N. Heat. Hey. Hey. Hey. Yeah, I think so. Very chill today. It's Sunday. Just going to enjoy a bit of vibe coding. Why not? Oh, look at that. Codeex has been shut down. Interesting. Absolutely. Nice. Yeah, Sundays are made for vibe. Cody, sure. Yeah, I think it just Apple picked up something and was like, "Right, let's delete that." Might have been an older version cuz I don't use it much. It is live indeed. Heat. Heat. Let's have a look. I think I'm on high. Yeah, let's have a look. Uh, extra. I'm going to crank that up actually to max. Should be max effort really. Heat. Heat. Heat. Heat. usage tier. Oh, yeah. Yeah, I'm on the high the highest package. Uh there's a lot of people building stuff like that inside the AR boardroom. There's a lot of post about if you search for that sort of stuff. Personally I don't do it but you can and there are a lot of people in the offer boarding who are doing that. So it's a good place to connect with people. Hey hey hey. Hey, hey hey. Daddy. Nah. Heat. Heat. Hey hey hey. Yeah, doing great, thanks. Tell you back. Today we are building out agents A1. We're testing that out. We got a claude AI SEO system. We're working on a new way to save 65% on tokens. A couple of free coders as well. We're looking at possible doing a Q&A as well. So that's what we're working on currently. Hermes aware Hermes or open code. Uh I think if you're just doing coding like ugly her uh not Hermes uh open code would be the best if you just want like an ID if you want an agent then you would use Hermes right so it depends what you want but Hermes is better for just scheduled tasks it's better for doing actual work it's not that great for for coding I think you better off with board or open code but depending on which API you use Yeah, we're working on that. No problem. Switch the music down for you now, sir. And then I'll switch you back up when I stop talking. Can you add a section inside there that says exactly how it works step by step, what the meaning of it is, how it works. Uh, right now it doesn't have that and it should explain exactly how it works. Yeah, I did a video on ponytail yesterday. It's pretty good. It's good for coding projects. Yeah, but literally all you've got is the actual instructions. Like you actually need to have the coder inside there and configure it for us. One thing I notic is you're not even using the guide that we've given you for creating the guide. So, you should be using the guide skills for creating guides, and you're obviously not doing that. So, make sure you learn that. Never heard of it to be honest with you. What does that stand for? Uh just started off by creating videos about SEO and then I used to do like one video a week and then just gradually ramped it up. I think honestly everything started to grow faster when I started creating videos about AI before that point. Like I never really got much traction at all. So I'd you know you want to f what I think works really well is just focusing on what's trending right now and then going from there. That's how you can reach a lot of people. But yeah, I mean everyone gets started in the same place, right? One step at a time. We actually got a six week course inside the air profit boardroom on on how to implement that step by step as well. Back in a sec, peeps Heat. Heat. Hey, hey hey. Heat. Heat. That's pretty cool. Like the idea of that. Daddy, daddy. Make sure That explanation is right at the top. Yeah, it's coming on the latest one. Don't worry. Bear in mind yesterday when we created that it was the 4th of July and so if you installed the 3rd of July it wouldn't have the fourth of July updates right? So the agent that's set up for you, you can ask them to to switch which model you're using inside Hermes. So if you're using a local model inside Hermes and you want to use that inside agent OS, then you ask the agent that set it up for you to use the local model inside agent OS. Yeah. So, I'd go with open code. Uh, it's a lot more polished, a lot more smooth. Hermes is good, but it's it's open source, right? So, it's going to be a bit buggy. It's going to be a bit messy. Uh, especially for coding tasks because it's more like a a genic model. You go back and forth with it, right? Um, so again, I wouldn't use I wouldn't use Hermes for coding. I'd use Claude code or open code. Daddy, daddy. And have you actually checked it's working and run some tests with it? So, for example, it says keyword researcher failed here. Make sure you've got it all working. And then also, when you're using it and making it work, add some screenshots and examples of that inside the guide. Yeah. Well, I mean the self-learning loop is is more just for agentic tasks again, right? So, I'll give you an example like if you were doing a research task where you were getting like the latest information from uh Twitter, something like that and you're trying to get like the latest new trending news. It'd be great at that, right? runs on the scheduled task, learns how to improve its research every day, learns how to find better sources and improve the organization of it. But if you're using Hermes for coding, well, there's no like real IDE. Um, there's no real preview of what you're building and it's not like it's more just for implementing stuff. That's the difference. So, but again, try it. See, try them both. See what you think. That's just my own experience. Can you make sure there's a workspace inside there, too? So today we're going to be answering some of the latest questions on the agent operating system that were built, how to build them, how to create them, how to get the most out of an agent OS, and some of the best features that we've built, plus how you can build your own and use your own to get the most out of this stuff. And the thing that I would say is once you set up an agent operating system, it literally changes everything that you do, right? I'll give you an example. Every time we have uh a task that is ongoing that I need to do like daily, I will literally just automate it as soon as I can and then build the custom workflow inside this setup. If for example, I see something pretty cool like for example, open montage as you can see here for generating AI uh films, then I just build it straight in. I'll show you another example. So this morning we saw an amazing uh operating system for SEO and it was so easy just to get that open source project and build it into this system as you can see right here so that we can have like a a real working office with our agents ready to go inside here. So if you're wondering okay like how do you get the most out of this stuff? How do you build them? How do you get started with all this stuff? Um what can you set up? Let me guide you through that today by answering some of the latest questions inside the air profit boardroom. Let's get into it. So, we got a question here which is, you know, I'm new here and I'm building my agent operating system step by step. My goal is to create a working setup for manage your projects, keeping memory and slowly connecting useful AI tools without getting overwhelmed. If you were starting again from the beginning, what would you focus on first? So, if I was starting from scratch again, I actually probably wouldn't use every single feature of this. Just focus on what works for you. So, for example, if you only use Claude, only build the Claude CLI into your agent operating system. Otherwise, what you're going to find is that it basically grows too big. So, I'd simplify it more. That's one of the first things. And then also, one sec, peeps. Why does it keep opening up this All right. How do you keep clawed code personalized to the ways you build? I see it's a context but the cross session persistence is it just a claw feature or something I can do myself in smaller proportion. So if you just want to keep the context you can just fork the conversation and build something else. That's usually what I do. So like if we're if if we build out a fe feature yesterday for example like uh a lead generation tool and then we want to keep building on top of that we can just fork the claw conversation. It still has the context as personalized to me. I'd also say look at what you're working on currently and then focus on that and automating that next. So whatever you spend a lot of time on, make sure that's the first task you automate cuz a lot of the stuff inside an agent operating system people build without actually using day-to-day. So you want to make sure it's all super relevant to you and personalize you. Then the final thing I would say is just make it your own. You know, like if you want to change the colors, the UI, if you want to, for example, add new features, make it your own. Make it feel like yours because the more you can personalize it to you, the more you're going to use it and the more useful it becomes. Next question from Rick. So Rick said, "I've installed the agent OS on VPS, signed up for news portal, configured N router, etc., but he's stuck on what to do next. Is there a complete checklist or step-by-step guide that runs through the full installation, configuration, API connections, and final testing?" So when you actually give the zip file to your agent to set it up, you can just ask it to install every part of the guide. So there's for example like 30 different modules inside the CLI section and you can just run through those and make sure everything's configured. But also note that you don't need to install everything inside the agents. Just pick the things that are relevant to you. You can always delete or ignore features because otherwise you might spend a lot of time connecting and configuring stuff that you don't actually need. Kenneth says avoiding boredom. This is a good one actually. So, I think a lot of people get into AI because it's like this shiny fun object that that seems really interesting and lots of cool stuff going on and there's always a new project which is amazing, but I know a lot of people are not productive even though they use claude code more than ever. And so, I think it's okay to be bored. Like, if you're bored and productive, that's totally fine. Honestly, the people who are the most entertained usually get the least done because they're distracted by shiny objects. They're distracted by new tools. They're not focused on automating the most essential parts of their business. So, if you're bored, but you're getting stuff done and you're making progress, that's the main thing. Well done. So, we got another question here from Sergio. He says, "Thanks for your amazing course in Agent OS. I've recently downloaded the Agent OS and was wondering if there are any script writer analytics loops or um uh social media content publishers, etc. Right. So inside there we have the agent video agent that you can use and the video agent is pretty good for social media content because it can generate video content for you across your platforms. for the actual publishing, I still get my team to post manually simply because it creates more accountability. And also, I know that some platforms don't like you to post with AI. If they see that, they tend to cap down on it and they tend to uh punish people for it. So, the way that I do it is I get my team to use these automations. They'll create the content, they can move 10 times faster, they can get way more done, but at the same time, they're still accountable to posting and keeping the quality high. And also this avoids like any problems that I see a lot of people have particularly for example on Twitter. I've seen people uh automate their Twitter. Twitter doesn't like that and then their account gets taken down. So we try to avoid doing stuff like that. So there's the video agent and then there's also the SEO section. So the SEO section can write blogs and content for you there. And then finally we have Hermes and Hermes Oracle. Now, Hermes Oracle, it can find topics that are trending from Twitter. It pulls those in. It gives you a new angle to create the content around and then you can draft the social media content or you can publish the content to WordPress. But either way, you can automate social media content using that too. So it basically takes in the latest news and then you can click on draft this and it drafts a piece of content for social media and also you can publish to your website too with the publish to WP option It's pretty interesting. So, Raj was talking about like an AI that can record and repeat repetitive skills. So, I think you can do this with Chrome browser use extension with Claude. So, if you click on Claude And then you click on teach Claude. There's an option to teach Claude. And then you can run the task inside your browser and it will teach Claude exactly how you do things by watching your screen and understanding how you do that. So you can teach Claude workflows using the Claude browser use extension. and Codeex has a a similar tool as well. So, we got a question from Jay here. Issues with paperclip. Some agents run, some agents don't. when I when I have these sort of issues. So, one of the things I find is like quite often paperclipip is a little bit buggy. I mean, it is an open source project, so that makes sense. What you can do in those situations is I would just get Claude to fix it for you and orchestrate it for you. So, what you can do is grab the paperclip GitHub details, plug that into Claude, screenshot and explain the issue you're having with paperclip, and then get Claude to debug it based on documentation and the error you've given it. That's usually the way I reverse engineer and fix it. and that's pretty much it for the questions so far. Now, if you're wondering, okay, what can an agent operating system do? Here's an example. So, we've got Hermes Astros here. And basically what this is is a version of Hermes agent we've custom created. It can pull in and analyze certain keywords and keep an eye on them. We can then see history over time. So we can see previous days and what it analyzed and then it will come up with new angles for that particular content. So if you look at this one for example, it analyzes the topic, then it gives us the titles and then from there it gives us the angles and we can plug that into our video agent, our SEO agent or even into notebook. And each one of these new topics, it gives us the same process, the same uh workflow. The other cool thing about this as well is that when we're looking at this, it organizes it in terms of what's trending the most. So, it's reverse engineering. Okay, based on all of these topics, which one got the most attention and then how can we target that, too. Now, on top of that, we also have Hermes Oracle. This can pull in the latest news from Twitter. We can see what's trending. We can draft SEO content for it. We can draft social media content for it. And we can see the original tweet, too. And then, additionally, we have the memory system here. Now the memory system basically analyzes what we've worked on recently. So our agents automatically update it and then inside the knowledge graph here we can plug this into our agents so they have instant context and every agent shares the same memory so they all know what we're working on as soon as we use them. Top of that we have for example this new SEO office system. Basically this is uh an office with our AI agents working together. And so we can ask each one of these agents questions. We can speak to them directly. We've got an inbox with them. Uh we can activate them as well, which is pretty cool. And then we can also run a technical SEO audit as well at any time and visualize our agents inside the same office, which is pretty crazy. So this was another open- source project that we plugged into the system. So if you want to get all of this, feel free to get it inside the AR profit boardroom. If you want to get my setup and also with the agent OS, you can ask questions inside the community, get help and support. I answer them personally with a video tutorial like this every day. You get the video tutorial here. You can see the last update date. You can get the full zip file. And we also add new tutorials based on what's actually useful. As you can see right here, so you get a video tutorial and a step-by-step guide on exactly how to use this stuff. Inside the community, you can ask questions, get help and support. That's where I personally answer all of the questions that were posted today. And then you can also get four weekly coaching calls where you get help support in real time. So feel free to get that link in the comments description or go to the aiprofitborn.com. Thanks for watching. Do you have Fable on a loop? Uh yeah. So for example, like this morning we installed a new local model called A1 and then we we tested it on a loop with uh like 45 different projects. So, we tested this with 45 different projects in the background, built loads of cool stuff out, and that just ran on a loop automatically. Uh, so that was pretty cool. Pretty simple and easy to do. And, you know, it can just run in the background whilst I'm doing other stuff. And that was with Fable 5. We got Fable 5 to install it and then test it. So and what you can do is you can uh you can use its dream feature which compacts like every day and then learns more and more. The other option that you have is that you can basically after every response train it as a skill to self-improve. So you can say okay every time you make a mistake document that using your self-improvement skill and if you don't have a self-improvement skill create it. And then every time you make a mistake, wire that mistake into the skill so that you don't make it again. Um, I guess you can already do that cuz you can go. So the question here is when will I be able to think of software and it will be made without even typing anything. So like if you if you just don't want to type, you can use whisper flow like that. And then if you just want to build like any sort of SAS software like for example we built the agent OS then you would just use um you would just ask it to build whatever your idea is and then tweak and test and implement it. Yeah. The only bottleneck is the operator itself. Exactly. We are close to automation which we don't need to audit the output anymore. I think it'll get there one day. for sure. Do you have the guide on it as well? Thank you. I I can't take credit for it. It's someone else's project. It's a open source project that we're looking at. can you please explain how agent factory and free claw code work together in your agent OS? All right. So the agent uh so look agent factory ah you're talking about this bit right so basically how this works is that I mean it's very simple we're using a free open source project called free clawed code and we plug that into the system then we can run it with local models so if we give a prompt here then we'll get a preview over here. So if we for example say okay build a snake game we'll start building it and then using a combination of free clawed code we've created this agent factory. Now the agent factory basically means that you can preview everything that you build in real time. So when we're using local agents like for example A1, we can see it coding out here. We can give it the prompt over here. We can control it with our voice over here and then we actually get the output over here. So that's basically how it works step by step. Now if you're wondering how to configure it, you can just use the documentation that we give inside the air profit boardroom and that shows you exactly how to set it up step by step. Um and you just get your agent to set that up for you. I wouldn't do it yourself. If that makes sense. Welcome. Happy Dub. Well, how do you get clawed code for free is a question from Andrew. So, it's a free open source project. Any updates on Deepseek? Already posted an update on Deepseek earlier this week about Deep Spec, which is uh a way to run local models faster. Agentic OS is a harness at the core but with just better features. So it's basically a way of just coordinating and orchestrating all your agents in a way that's more fun, more interesting, more user friendly, etc. Today I'm going to be looking at a brand new free local model. It's called Agents A1 from China. This is uh from a lab in Shanghai I believe. Basically, this is a free local model. It's a vision language model built for aentic tasks combining multimodal understanding with tool use. Basically, this means that you can run your agents free forever and you can code whatever you want. And I'm going to show you exactly how it works step by step. We actually built out 42 different things with it on Goldiebench. So, I'll see I'll show you how it performs in a second. And what you can see here is that you can run this with free clawed code and we've got an agent factory which means basically we can give it a prompt like for example build me a snake game. It will go off and build it with HTML in real time which we can preview and then we'll get the preview on the right hand side in terms of what it's actually built and we can see everything that we built previously over here. And so we have a free local model that can actually build out cool useful stuff and you know it doesn't cost anything to run which is pretty nice. that actually works. Like this is pretty nice. Uh, interestingly, if we have a look at Goldie Bench and we look at the leaderboard for local models, if we have a look at the local leaderboard, it's scoring number two right now in terms of the benchmarks here. And we've come we every time I test a new model, we run it with like 42 to 45 different projects and you can see how it's performing here, plus how big the model is. Uh, you know, these are all open source local models, how we're running it, and then also the details of that agent. So this is an agent tuned open mixture of agents model built for long horizon tool work running free on your Mac right uh or on your you know it doesn't have to be on your Mac it can be wherever you want it to be so it could run on on uh PC as well and you can see the details here so it's a 35 billion mixture of expert agentic model it's pretty cool it's also supposed to be really good at science although I mean I'm probably not the right person to test it for science so I'm not going to do that myself but you can see how it performs on the benchmarks do I listen to benchmarks, especially with local models, absolutely not. That is why we created uh Goldiebench. But if you do want to see how it performs versus other local models, you can see that right here. So, uh this is versus, for example, step 3.5 flash 3.6. Apparently, it's beating it on the benchmarks. Test out for yourself as I always do, and then you can go from there. So, if you're wondering, okay, what is this good for? What is it bad for? We'll come on to that in a second. If you want to see what we actually created, then you can see some of the examples here. So, for example, this was like a open world game that we built out. Um, and it actually worked. I was surprised like most local models totally fail at the task and give it, you know, give back something terrible, but this open world game actually works. It's a little bit buggy in parts, but overall better than when I've tested it. Bear in mind, like usually local models are not great for visual tasks, but this seems to do pretty well. got like this neon racer game. A cool kind of synth wave style racing game as well here. So, you can actually build stuff. And this just ran in the background whilst I was coding loads other stuff. So, it seems to be more lightweight than most models I've used, especially because I'm running this on a Mac Studio and it it seems to work okay. Usually Max Studio with local models doesn't work so well, but it's getting better and better. This was a a landing page that we created with it, which is not bad. So, you can use it for building websites and that sort of thing. And you could also use it inside an agent operating system. So for example, you could use it with the local setup over here. You could use it with free claw code. You could also create a new profile with Hermes agent using agent01, sorry, agent one and then plug it in and run a genic models for free as well. So it's 35 billion parameters but it only wakes 3 billion at a time. That's how it runs at a small model speed but still gets good outputs. They also tuned it in three stages. So this is from intern science and they tuned it with supervised fine-tuning domain teachers multi-teer distillation and it's specifically designed for agentic work. So long horizon search engineering instruction following etc. And the total context window of this is 256k context in terms of speed if you compare it versus other models. So we compared it against Gemma 4 12B MOX Gemma 4 Koda Quable and it's performing the best out of all of them. It's the fastest. So it scores 95 on A1. And if you want to see okay what can it build in real time we can say for example okay build a landing page for an SEO agency and then we just click on build over here and we can actually get a preview in real time in terms of where it performed. So Quael 27B coder is still outperforming agents A1 by quite a long way but it's outperforming Gemma 4 Laguna XS 2.1 and Quast as well. Now, if you want to see some example builds, Here we go. So, it created like this game that was pretty fun to play. We've got the open world dragon game. We have the website landing page and we have this car game as well. again like when you're building stuff with local models typically the outputs are going to be like using Opus from a year ago if that makes sense. So it's never going to be like okay top tier frontier quality but if you just want to see what it can do those were some examples. The way that we test it as well is we basically give it like one prompt A1 builds it will get the actual output and then Opus will judge it and it runs for free offline. So if you don't have Wi-Fi you can still run it. If you're on a plane, you can still use it. And you can also wire it into the agent operating system. You can see here when it's coding out, it's pretty fast. Also, from what I can see, it's not available on OAMA just yet. So, that's one thing to be aware of. By the way, this is the page that is created. It looks pretty nice, right? So, we asked it build a landing page for an SEO agency. It's got a nice little background here. Looks quite clean. for a oneshot prompt. That was pretty good. So, I think it's great for like just coding out and also the basic tasks, right? So, content writing, blog posts, creating and coding out websites. It's fast, it's free, actually looks nice, it's decent at UI, etc. So, not bad at all. Um, this is running on a Mac Studio Apple M4 Max. 36 GB seems to work pretty nicely. And if you're wondering about A1 where it comes from. So, according to what I can find, and there's not that much information about them, but they basically come from Shanghai. This was released on June the 29th. They've got a full paper talking about it and how it works, etc. And apparently it can reach trillion parameter level performance. But yeah, overall pretty good, free, local, open source. Now if you want to get the full system for coding with this using uh free claw code the local system that we set up plugging into Hermes etc you can get that inside our agent operating system which we give away inside the arprofit boarding link in the comments description or go to the arrum.com and inside this community you can learn how to scale and grow with AI automation inside the classroom if you want the agent OS with our best local models plugged in you can get that right here there's video tutorial you can see the last update date you can get the zip file to use it And then you can plug any free local model into the system to build and code whatever you want. We've also set up recently Omniroot, which is another way to code for free with AI and automatically roots across 90 free providers. So you don't need an API key for that as well, which is pretty cool. Also, we add new daily tutorials like you can see right here. Inside the community, I personally answer all the questions and so does the rest of the community to help each other as much as we can. There's always people online 24/7 to help you. And inside the classroom, you get access to all of our best trainings. Inside the calendar, you can jump on weekly coaching calls, get help and support in real time. And in the map, you can meet people in your local area who are building out AI agents with local and agent operating systems. So feel free to get that link in the comments description or go to the aprofit.com. You also might say, okay, how does this perform against orif, right? Or is another model that a lot of people are liking at the minute. So, for example, if we have a look at that dragon realm game that we're talking about before, on if 1.0 doesn't didn't seem to get past the start screen. Whereas, for example, when we were testing that out with agent A1, it could actually create something half decent. half decent. I wouldn't say it's like it's definitely not production ready, but you get the point. So, I mean, if I had a choice between them, I would go with Agents A1. It seems faster. It seems to create better outputs. You saw on the website it great as well. It's pretty nice. So, yeah, that's my opinion. Thanks for watching. What AI do you recommend for studying? Probably Claude. I would just go with Claude or notebook is pretty good, too. on if I already covered that 35 GB. Uh no, no, no. So it's 35 billion parameters. The actual size of the model is 21 gigs. Regent ask how good is it for tool calling. It might be something we test out later. I hope GLM 5.3 will be released soon. Yeah, I mean GM 5.2 came out around the 17th of June, so I can imagine there's probably a new version that will come out this month. Hi, I'm new trying to code with AI. Just starting to get a grip on what works best. Just got max and can almost run the 5hour limit now. Nice. Have you built your own editing software? What do you mean? Like for videos or for what? For videos. Yeah, we have the video agent inside the agent OS. So, if you go down to the video agent section here, you'll find it. Are you running Fable 5 right now? I am indeed. There's Fable 5. Yeah. For videos, we've got the um we've got Remotion plugged into the agent OS so we can edit Yeah, I mean so that that is basically software, right? Like if you look at this section here with the video agent, you type in your prompt, you get back what you want. This is a fully custommade tool. You can't find it anywhere else. That is that is basically software uh plugged into the agent operating system. So, we have this video editing tool that is plugged inside the agent OS and then we can just type the prompt and and get the outputs from there. Today we're going to be looking at a new open-source project called SEO Office. And this is a way of having a team of agents so that you can actually see what they're working on right here. And you can also click on any of these agents and start running them. So you can see for example we can pull up the technical SEO auditor. We can speak to it directly and we can give it new tasks. Now you've got the inbox and the conversation and then you can speak to it directly which is pretty cool as well. And this is the same for every single agent that's inside the office. So we also have for example this SEO schema agent and we can speak to it ask it to do stuff ask it to work on our projects etc. And so you can have a full team of agents. You can even see what they're working on directly on the screen which is pretty amazing. It's such a it's such a cool way of visualizing your agents and just seeing what they're working on. Um, and I think it it's pretty interesting. And then we've got the main one in the middle, the main man over here as well. Now, what we can also do is we can click on run technical audit and that actually start running a technical audit on our websites as you can see right here. So, it's going to start running in the background and then we can see which agents we've got. So for example the backlink anal an analyst the keyword researcher the technical auditor etc and you can see what they're working on you can see how they're working step by step etc. Now this is based on an awesome open-source project from Daniel Agreei. So it's using a combination of claw 3D UI claio specialists and marketing brain. So, three different projects plugged in to this system. It looks absolutely awesome. And then we can speak to our agents too. So this is our main orchestrator. And when we're chatting with it, we can also speak to it directly here. So we can select which one we want to use and then speak to it directly. So if we look at the orchestrator and we're like okay create five SEO optimized articles for the keyword Hermes agent skills and then we can select whether we want auto readon plan mode or full access and from there we just click on send and we can start using the orchestrator and you can see it's now thinking and we can move or change the size of the chat window as well. So, we're going to have the chat window at the top here. And we can see our agent just working and getting stuff done. And now it's actually replying and it's dispatched keyword research on home agent skills. Then it's going to topic cluster to identify five distinct article angles. then content briefs for each and then it gives you a SEO optimized article from We can also minimize the chat if we want to as well and then reopen it and we can see what each of our agents are working on too. So we can see the orchestrate over here. But then if we zoom out, we can have a look and see, okay, what what's our keyword researcher working on as well? So let's have a look and see how it works. So number one, it's local first by design. So everything that you do inside SEO office gets saved locally. It works in parallel. So you can dispatch up to 25 specialists at the same time, which is pretty cool. So your orchestrator can spawn specialist agents. You can see for example here the it's finished the orchestrator dispatch and then it's spawning specialized agents. You basically get updates as it goes along and as it keeps working etc. Also, it does say inside the notes here that you can clone it, you can run it, you can fork it, you can build on it. You just have to keep the network server modifications open. And here's what it looks like mid audit. So the brain is a per client second brain. So everything that's created inside the system gets plugged into a second brain that's visualized like that. Then you have an orchestrator that runs the whole project. So that's basically like the agency project manager. So if you look at this whole system, it's basically like you're replicating an SEO agency and then you can have different memories and a second brain for each client and the orchestrator will delegate each of the tasks inside the project to different sub aents. So let's say for example you're like an SEO agency, you know, you could run multiple client sites for this. If you're inhouse team lead, you could use this or a freelance SEO client as well. Now you might also wonder how does it work step by step. So basically this is the open source project and then you can plug in your data for SEO API key and that runs through the whole system. Now we also have open SEO plugged into our system and we have a full set of agents for doing the research from your Google search console, generating content, deploying it, etc. So when you're using something like open SEO, you can combine it here and then you can see what your agents working on. And this is the second brain by the way. So you can see here that it's fully visualized in terms of what's being worked on and what's inside the notes and etc. Right? So if you click on each of these, these are different memories as you can see here. So like the seasonal keyword playbook, the location guide template, these seem to be like different skills that are automatically in installed. And then depending on what the orchestrator agent is given as a task, it will use one of these different playbooks for doing SEO, which is pretty nice. So to get it set up, you can just use the GitHub. Then you use the data for SEO API key. You load it up and we put it inside a agent OS as an example here. Um, you can do whatever you want with it, but I mean like it could sit alongside for example how you deploy the content or for example how you do the research. So you can actually like look at your Google search console here, come up with keyword ideas, plug that into SEO office, get the orchestrator to create the content and ideas, and then based on the content it creates, you can deploy that to your websites using this tool. So it kind of works together in a nice workflow. And then also any audits you do are automatically saved inside the wiki. And just as an example, we actually did a backlink audit using the data for SEO API key here. So you can do like a backlink audit that will look at your numbers on your website. You'll audit our website for example like Julian.com, give us a summary, the data, the referring domains, etc. And we can also use that for keyword research as well, which is pretty nice. So, it's kind of like your own customizable SEO agency that can orchestrate teams of agents, but specifically for SEO and creating content that way. Pretty amazing stuff. Now, if you want to get our full agent operating system, the SEO office itself is free and open source. You can get it from GitHub. If you want our full SEO tools with the research tool, open SEO, the content deployment system. Here we also have Hermes Oracle for finding trending topics and then creating SEO content around that. And we have Hermes Astros that can actually find trending topics give you unique angles and then you can plug that into your video agent or your SEO agent as well. So, however you want to use it, feel free to get that inside the AI profit boardroom link in the comments description or go to the aprofit.com. Inside the community, you can ask questions, get help and support whenever you need to. I personally answer every question in there with a daily video tutorial. Inside the classroom, you can also get access to all of our best trainings, best lessons, etc. Inside the calendar, you can join our weekly coaching calls. We have four of those per week where you can share your screen, ask questions in real time. And if you want to get the agent OS system, you can get that over here. And we update it daily with video tutorial plus a zip file to install and you can get new daily updates like so. Inside the map, you can meet people in your local area who are building with AI agents like you. That's all available inside the arpuffing link in the comments description or go to the arpuffwarm.com. Did you open source that? So, it's not the SEO office is not my project. It's from Daniel. I think I pronounced his name right. I think it's Daniel Agrii. But yeah, just go on to GitHub and look for SEO office. Just joined. I'm a new user of claude code. Working on multiple projects one by one. How to develop multiple projects efficiently. Yeah. See, I mean, you don't need to do it one by one, right? If you want to work on different projects at once, you can use Claude desktop like this. And you basically switch between conversations. So, for example, when I'm creating like multiple different features inside Claude for for different projects, I can just open up a new conversation and have my agents work in parallel on different projects. So, that's what I recommend for you. I think that's probably the easiest and best way to do it. Joan says, "That looks awesome. I like the visual illustration of each agent. It helps people visually understand the concept of a digital employee 100%. But yeah, it looks really cool. Jungle Build says, "Whoa, no music today. We had some music on earlier, but I'm just uh recording some tutorials, so we'll add that back later. So today we're going to be looking at a new way to cut 65% of tokens with clawed code. So if you're using Fable 5, this is an easy way to cut the amount of tokens you use, but still get the most out of it. So you could use it even more. I mean imagine if you can code 65% more with Fable 5. How much more would you get done? 65% more. So that's why we are using this system to reduce the amount of tokens that we use when we are running AI agents. And you can also use this with any other agent. You know, it's not just for Claude code. You can also run it with Codeex, Gemini Cursor, Windsurf Klein, C-Piler. And it's free and ready to go as you can see right here. You also might say, how does this work? Well, I'll run through that in a second. And if you're wondering, okay, how to install it, you can just copy and paste this command and give the whole setup to Claude Code so that it can run with Fable 5. And the great thing about this is you can install it for one agent or even for 30 different agents, which is pretty insane. And the way this works is basically you talk to your agent like a caveman. Um, and when we actually tested it on Fable 5, it used 69% fewer output tokens. I'll show you the example test in a minute, it was 37% cheaper, and every answer was still pretty good. So, here's the problem with, for example, you know, something like Fable 5. It uses up a lot of tokens. When it stops being free on the subscription and you have to pay per API, it's going to get quite expensive if you're not careful. And so, Caveman is a free little skill that fixes the talking, not the thinking. And you install it once and then your agent basically answers like a smart caveman. Short, blunt, and right. And then the way this works is basically, you know, the the idea is that when an AI answers you, you are using up tokens for every word it says. And those are output tokens. And on the smartest models, they're quite expensive. They use up a lot of tokens. With Caveman, it's a set of instructions your agent reads once at the start of every chat. And the instructions say, for example, drop the filler words. Uh, no, like for example, sure, I'd be happy to help. No, basically and actually, no three paragraph warm-ups. You say the thing, you give the fix, and it stops talking. And there's basically three rules inside Caveman to make it use less tokens. So number one is code is sacred, right? So code blocks, commands, file paths, error messages are never touched. Then also it knows when to speak up. So obviously sometimes it will need to speak more in situations related to security and so this way it knows okay during those situations we're going to speak more and also it can speak multiple different languages as well. If you for example write in Portuguese it will reply like a caveman in Portuguese. It compresses the style but not the actual meaning. And so the way this works is like it thinks exactly the same. It's still thinking like it would normally but the difference is the filter afterwards. So all the filler words that come back to you um are reduced and so you use less output tokens when it's replying back to you when it's coming back to you inside the chat. Also you might say okay you know aren't shorter answers when it's replying to you like making it worse. Well basically apparently sometimes this can be the opposite. So, a March 2026 research paper tested 31 models and found forcing brief answers improved accuracy by up to 26 points on some benchmarks. And in my own fivep prompt test, every caveman answer contained the same fix as the long ver version, right? So, it was still working just as well. It's just like it replies with less words and it's a bit shorter and more straight to the point. And if you're wondering, okay, what it actually is and and what happens in every message, basically, it's just a text file full of rules. So that's a whole secret. It's not like an app or new model or a middleman or anything like that. It's just like a written instruction, a skill that tells your agent how to speak, nothing more. Now, the installer basically copies that file into your agent. So you run the one line install, it looks your machine, finds every agent you have and then with claude code, C codeex, Gemini, 30 others, it drops the same rules file into each one skill or plug-in folder in the format that agent um expects. And then also your agent reads the rules at the start of every chat. So AI agents always read their instruction files before answering you. That's how skills work. So from message one, the caveman rules are sitting in the model's context right next to your question. Now they also the rules tell it like what to stop. So the actual file says almost word for word drop articles, drop the filler, um drop pleasantries. So it's not going to say like happy to help sir, thank you etc. And also it answers in the pattern which is thing action reason next step. Also the rules tell it what to protect like we're talking about before. And the model writes short and it never edits answer afterwards. So it's not going to like trim the answer. it just responds in a short way straight off the bat. So here's an example. If you ask like, okay, why is my React component rdering? The agent bundles your question in the caveman rules file. The model reads the rules. So it thinks exactly the same way, just chooses fewer words as it writes and you get a short reply. So instead of getting, for example, like 1,349 tokens used on the answer for the output of this, it would use like 324 tokens in our test. So, we actually ran this with five different tests using Fable 5, and it's pretty interesting in terms of the amount of tokens saved. On every single one, we saved a lot of tokens. Um, we actually ran the same five real questions from the GitHub directly to Able 5 using the caveman skill. It was pretty interesting. So, you know, the normal out, for example, for like a React rerender bug, that would be normally like 1,349 tokens, but with this, it was 324. And you can see here, like on the five tests, we saved a lot of tokens on every single one. The average was about 69% saved, which is pretty insane. So 69% fewer output tokens, which would save about 37% of the total, if that makes sense. Cuz this is just for output. It's not for input. And you can use it and install it with one single command, which is this one right here. You could also use it inside your agent operating system. So, for example, if we were using Claude, and this is actually something that's really, really good is like because you've got so many agents plugged into the agent operating system, if you're using multiple agents like Hermes, like OpenClaw, like Claude, like anti-gravity, you can install caveman across all of them. And then it's reducing the amount of tokens across all of them. And so like it has a massive compound effect because you're saving 69% not just across one CLI but across every single agent and CLI that you use which is pretty amazing in itself. And we checked the answers were right. So all of the answers were much shorter as you can see but also they were all correct. So the great thing is you actually get the right answer back. Now this really matters on Fable 5 because obviously number one you've got limited tokens on the subscription plan. Number two, when the subscription plan on the 7th of July drops, well, at that point, you won't be able to use Fable 5 on the subscription. You have to use the API. If you're using the API, you want to reduce the amount that you use the API, otherwise it can get expensive. So, this way, you can just shave off a lot of tokens and save a lot. So, let me show you like the old way versus the new way of this. Like the old way would be, for example, every reply opens with like, "Sure, I'd be happy to help with that." or something like that, right? And then you got three paragraphs to set up before the actual fix. You scroll and skim to find the one line that actually matters. We've all seen that inside, for example, Claude as well. Like you can see quite often it will give you a long answer like this, but you can reduce it massively by using caveman. And then also the great thing about this as well is like quite often in memory files when you're using clawed code and that sort of thing the replies are trained to be quite buggy or quite full of too many rules if that makes sense. Whereas with this system it uses 69% fewer tokens on my own test with Fable 5. You read the reply in one glance which is faster than skimming because it's much shorter. Everything stays intact. So all the code, the commands, everything else you have and then the memory files compressed once, which means that once it's compressed, the savings repeat and it's one free install, right? It's a free and open source project. And you might also say like, I'm not technical. Can I actually install this? Is literally one command. So if we scroll up, all you need to do is use that. And if you don't know how to do that, you can just give the command to get uh you take the GitHub and then give it to Claude code and that can set up for you as well. Now you can also switch this off at any time too. So you can do forward slashcaveman to bring it back at any point. And you can also pick how it replies. So you can switch between light, full, and ultra. So if you don't want to use this all the time, no problem. If you want to use like some of the bonus commands like caveman commit, for example, or for example, caveman review or caveman compress, these are all different options. You can reduce the token count even more as well. And you can see how the responses vary. So if we look at the system, if you're using like the normal, it would reply like this. If you're using light, it would reply like this. With full, like so. And with ultra, even shorter. So you can make the response shorter depending on which setting you have. If you're wondering how to change the setting, you just change it like so. So that's basically it. That is how I saved 69% of my tokens on Fable 5 using this system. Pretty powerful. If you want the full token saving stack, caveman shrinks and replies, the agent operating system is designed for that as well. So it can work with all of this. The other cool thing is that we have free local models inside the agent operating system. So you can use, for example, the local system here. And we also have Omniroot plugged in as well, which means that you can code for free using a system that delegates everything to free APIs as well. On top of that, we have loads of cool features inside the agent OS like Hermes with Hermes Oracle, Hermes Astros, a voice agent called Apollo. We have Claude that can code directly and we have a memory system plugged in as well. So this is a powerful system to make the most of Caveman so that you can use even more tools but reduce the amount of tokens. So we have loads of token efficiency playbooks as well inside the AI profitable border. So if you want to get that feel free to get it inside the AI profitable room link in the comments description or go to a profitable.com inside the community. You can ask questions and get help and support in real time and I personally answer every single question inside there with a video tutorial inside the classroom. You can get access to all of my best trainings including the agent OS system that we update daily. You get video tutorial full guide on it and the zip file to install it. We also add new daily guides based on what's actually useful that has just dropped and I test all of this stuff personally as you've seen today. So for example, today we tested out five different experiments. We do that with every single tutorial inside the AI profitable room. You get the video tutorial, the full step-by-step guide and everything you need to win with this stuff. Inside the calendar, you can drop a weekly coaching call, share a screen, ask questions in real time. Inside the map, you can meet people in your local area who are building with AI agents like you and you can connect with me personally. So feel free to get that link in the comments description or go to the aipruffborn.com. Thanks for watching. Thanks. Yeah, I might put that on the list and check it out in the future. Let me put it on my list. Here we go. We're looking at a new system called Omnirroot today that allows you to code for free forever with AI and I'm going to show you exactly how it works step by step. We just built this out as you can see. So we said animated starfield on a canvas actually built out perfectly first time round. Gives us the code gives us a preview and this is running through a free API. Now the cool thing about this is basically you have one local endpoint that roots for example claude code cursor um across 237 different providers and 90 of them are free forever. Plus it also um falls back when you run out of tokens and there's token um minimization systems built into this that reduce the amount of tokens you use by 15 to 95%. And this means you can code for free forever with AI. Pretty insane. So this is a free open source project. It's called Omniroot. It is a free AI gateway. Now what this means essentially is that you never have to worry about stopping coding. You can connect every AI tool to 237 providers. And you can reduce the amount of tokens use each time you run with this as well, which is pretty wild in itself. Now this also runs in 42 different languages and it was super easy to set up to get it set up. I mean, I've already plugged it into the agent OS. We created a new tab called Omniroot and essentially I just gave the GitHub to Claude. Claude set it up for us, plugged it into the agent OS, as you can see right here, and it's ready to go. We can also save stuff that we built and then we can come back to what we've built previously as well. Open those up, have a look at them in real time. Let's test it again here because it it seems to work really nicely. And also, we've got instructions here on how to install it. But literally, if you want to install it, it's really really simple. So you just use this terminal command set up and then use this terminal command to point claude code to it. So what that means essentially is you can run omni root with claude code and that means claude is free as well. Pretty impressive in itself. So if we say okay you know just build a to-do list app as an example. We'll click on build and that will start routting through free providers and then with those free providers we can start getting outpost directly here as well. So this is something that I call the free coding engine. You know, I think a lot of people sort of run through the problem of like, you know, they're using Claude, they're using cursor, they've got other coding tools on the side, and then the free tiers run out in the middle of a build, one provider goes down, the whole day stops without it, or you have to switch everything over, and that takes time, too. And so, if you're looking like a free way to use all your tools without worrying about this, I mean, you can see the the build here. It looks pretty good, you can just use Omniroot instead. It's way simpler and way easier to use. And then if we want to save anything, we've actually created a custom build around this. So what this means essentially is like we're using the power of omni root. We've got it inside this engine that helps you visualize and save the builds and everything else. And that's inside our agent operating system. The other cool thing that we can do with this because we built it in the agent OS is like we can plug it into our memory. We could plug it into Hermes. We could use it however we want. But the main thing here is like it's is pretty wild. Now, in terms of the benefits of this, well, you don't hit limits. You save tokens. It's $0. You got one endpoint. Every tool works. And the quality is different as well. So, how does this work? Well, it works in five layers. So, you ask, omnirute picks and falls back. You got 93 providers and then you get the build. And there's one door here. So every coding tool points at a single local URL. Then it uses a pool of 93 providers. It auto fall backs. So the reason for this is like sometimes you hit a limit on a free coder and this way it just switches automatically. It reduce the amount of tokens to use and it runs locally with a local endpoint. You also might say okay well free models they're not that good. But actually you can see two different builds that we've done with this. So the to-do list app and also this example as well and it looks pretty good. I mean let's test it again. The other thing that I was surprised by was how fast it is. So, if we say, "Okay, build a landing page for an SEO agency." And it will start running in the background right there. Let's leave that on the side so we can see what it does in a second. Now if you want to install it you can code for free in about 5 minutes. So you just use this command then you can point and claw code at it or you can use it in one click inside our agent operating system inside the air profit boardroom. And also because it's so simple to set up you don't need to be technical to use this. And so you've got one endpoint that can route to multiple different free providers. And then we've got the code back as well over here, which is pretty good. Also, what would be interesting here is if you could run it inside Hermes. So if you use a custom endpoint, I wonder if you could run the model directly via Hermes locally. That might be something we look at as well. And then you could run Hermes locally with this setup as well. Bear in mind, Omni Route is not like a local API, but it's a local endpoint. And that's how it works step by step. Now, you also might say, well, what if one of these free providers just disappears? And for example, if you look at our alpha or any of the free stealth models on open router, eventually once they come out of beta, they're no longer free. And so the great thing about this is that the whole point of omni router is it expects these three models to disappear eventually. When that happens, it auto falls back to the next model that it's lined up. And that works perfectly as well. As you can see this keyboard app that's ready to go here. Pretty nice. So I mean I've tested on three different things here. The virtual keyboard, the Tazulus app, the animated starfield works for pretty much everything. Runs across 92 different models. You can also use varants. So for example, you got auto, but then you can also use like auto coding. So depending on what you're doing, you can switch the model. So if you just need something really fast, you can switch to auto forward/fast. If you need something good for coding, auto/coding offline, you can run with that. And then there are 17 different routting strategies. So for example, you could go with the priority, you could go with weighted headroom or even fusion as well. So this is pretty interesting because fusion is basically a way of using like multiple different AI models at a time and then fanning out the task to multiple different models and a judge will synthesize the answer. Now you can actually use fusion as a rooting model with omniroot as well. So it's pretty exciting. and I might keep testing it. Be interested to know what other people are using, but it's a free way to code forever with AI. What's not to like? All right, so if you want my full setup with the agent operating system and the free coding engine plugged in, the free coding engine is one tab. The agent OS is a whole room. Omniroot is the free coding backbone. But inside the air profit war you also get the entire Asian operating system with the omniroot free coder Asian cambban every CLI you already pay for the AI mastermind the local hermes engine the claude workspace uh free local models etc. We've plugged that all into the agent operating system and inside there you also get an amazing community of people learning growing and helping each other with AI automation. There's 3,900 members inside there which means that we can all help each other. There's always people online 24/7 and I personally answer the questions inside the community. Plus, everyone else helps each other too inside the classroom. You get access to all of our best trainings. And if you want the full agent operating system, you can get that right here with a video tutorial, the last update date and zip file to install it. And then we add new tutorials as you can see based on what's actually useful. Plus, inside the calendar, you can drop off four weekly coaching calls and get help and support in real time. Inside the map, you can meet people in your local area who are also building with AI agents. And that's all available inside the AI profit boardroom. Link in the comments description or just go to the aiprofit.com to get access. Thanks for watching. Hawk says, "I don't do SEO work, but you've given me massive inspiration for the agent OS I'm currently making and just watching those videos helping me avoid some pitfalls." Oh, awesome. Well, I'm glad it helps and really appreciate you watching and being inspired by it. I'm excited to see what you build. Today we're going to be looking at this new trend in GitHub that gives you a list of free APIs so that you can learn and build with AI free forever. So what this does, it breaks down all of the latest free providers and then you get a full list of like everything that you can use for free. So for example, we have a look at open router and you're wondering okay like what what models could I use for free with open router? Well, there's actually a list of all of them right here which is super useful. So for example, if we go to this model right here, it actually links directly to it inside open router which is super useful. So for example, we can go to Hermes and there's a few ways you could use this free model, but the main point is you could plug it into your agents, you could plug it into free clawed code, you could use the API, you could even, for example, start using it directly inside the chat here. So the great thing about this is, you know, a lot of people they don't have the resources or they're using a lot of tasks with AI agents. They want to keep the cost down. This is a great way to just reduce that. So if we have a look at this model, you might be wondering, okay, well, how do you start using it? Well, number one, you can go inside the playground here and you can start typing and talking directly to it. Or what you can actually do is you could plug this into your agent operating system as well. So if we have a look at this model, Hermes 34 5B instruct. If you go inside Hermes and we type Hermes model inside the terminal here, we can switch the model provider to open router. Plug in your API key and then you can select the free APIs you want to use as you can see right here. Okay, so we've got three diff three different free APIs that we've already plugged in to open router or what we can actually do is we can enter a custom model name, grab the details of that model, plug it in here and now the default model has been changed. So now next time we use Hermes, the model is changed to the free one. and we can go from there. Pretty amazing stuff. So you can use this model directly on open router. You can use the API key inside your coding models. The other thing that you could do for example is you could go inside free clawed code. So we got free clawed code plugged in here and we could use a free API directly with free clawed code inside our agent factory. We've currently got a free local model called agency A1, but we could actually switch the API to Hermes or whatever we want to use as well. Previously, one thing to know, just want to be 100 transparent with you, is that you see here it says Hermes 3 and four models are not agentic. So, if you are using models that are free like this, just make sure that you select an agentic one that's designed for tool use. Otherwise, the model won't work that well. So, for example, like A1 itself is designed for agents. It's a free local model you can use. It's pretty fast and easy to reply and it can build pretty nice stuff as you can see right here. But the main point here is like there's many different ways up the mountain to use free APIs and you've got an awesome setup where you can switch between them. So, there's a bunch of others on here. So, for example, open router. Each one of these will link directly to the free model. And then there's like different setups. So for example, you've got open router, you got Google AI studio with a list of free APIs. We have for example Nvidia, Mistral, and these are all free providers. Now on top of that, what we've also got broken down here is providers with trials. So these are kind of like ones where you can just test out, see what you think. Pretty reasonable to test out and then from there you can always cancel later if you want to. Um, there's three different tiers here. You got three tiers with the cloud APIs. You have the trials that we talked about a minute ago, and then we also have local models. So, for example, I've shown you A1 before. We can build stuff with these local models, and they're pretty decent. Like I've tested out loads of different local models inside our Goldie Bench leaderboard. So far, Quable is winning out of all of them, but Agents A1 was pretty decent when we tested as well. It can build some cool stuff. And you can see everything that we've created on the website, too. So, if we go back to the local engine over here, we've got agents A1 ready to go. We can see the stuff that we built. We can check out our workspace if everything's saved there so we don't lose anything. We can preview it, open it or download the HTML and then also we can use it inside the chat. So it's kind of like using chat GPT but with free local models that are ready to go whenever we need them. So if we say okay build our landing page for an AI community called the AI profit boarding that will start thinking locally and coding it out. And then once it's finished thinking locally, we can preview it. Now the good thing about local models versus using an API. So for example, you saw Hermes earlier, which is an API. When you're using an API, if you're working offline, then you can't use it. If you don't have Wi-Fi, you can't use it. If you're on a plane, you can't use it. But with local models, you can. And also a big difference between the cloud API and the local is that with local, your data stays on your computer. Whereas, for example, if you're using a cloud API, your data will go to the cloud. And now you can see it's coding out directly for us whilst I'm talking to you. And it's pretty simple and easy. It took a couple of minutes just to warm up and respond, which is pretty normal for local models, especially depending on the setup. So, I'm running this on a Mac. Uh, another good tip is that you can run MLX, which is a free open source project, and you can run local models with MLX and get even faster and better outputs. So, for example, if we look at some of the recent models that came out that we've tested, Gemma 4 on MOX now responds 90% faster. And you can run that with Alama too, which means like you can build out more stuff and you can build out faster, but you still get the benefit of a free local model. Now also if you're wondering okay like how can you is there a way like you can automatically route between them. What I've seen recently is you can use something called omni root. Now, when you use omni root, what that means is that it has a list of all of these model providers like you saw with the GitHub we looked at a second ago. But with Omniroot, basically you get a local endpoint, then it automatically roots the tools you already use across 90 different free providers. So, for example, if we go back in the agent OS here and we go down to the Omniroot section, this can basically code for us. We can again see everything that we've built inside our workspace here and we just chat with it here and it roots automatically. Now the good thing about that is for example if you're using a cloud API like for example Hermes before if that gets rate limited then you're kind of stuck unless you can quickly switch to another one and it takes time to switch to a different model or to reload the model or to change the API settings. Whereas with omniroot you get one local endpoint and then it automatically switches and falls back to different models depending on what's rate limited and what you can keep on using which is even better. So that's kind of like a way to automate the free models that we talked about before. Now you can also use something called open routter colon free and that way it's one API key but there's dozens of free models behind it. So what this does you give it one key then it has a router and then it switches automatically to the most relevant best model. So if we go to open rout here and we type in free models router. This is the model that I'm talking about. And again, you can use it inside the chat. You could plug it into free claw code. You could get an API key and then give it to Hermes. However you want to do it. And you see the top models here. So, for example, like GPSS, Neatron 3 Nano, Neatron 3 Super, Neatron 3 uh 9B V2. All pretty good options. Now also we were talking about agent um agentic free models before as well and that's important to note because the thing to note here is like as you saw in the warning on Hermes before if you're using like Hermes and some other different models they're not agentic they're not designed to be used with AI agents like Hermes or for example openclaw because they can't call tools so what you can do instead is you can use an agentic API that's still free for example like damage run 3 ultra and if you Look at the descriptions of these three models. You'll know which ones are agentic. So you see how that says Neatron 3 Super is designed for complex multi-agent applications. That would be a more relevant better model for AI agents. So depending on what you're doing and what you're working on, you would switch between them. Now also what's pretty cool with omniroot is that depending on the task you can actually set the type of coding you want to do and that will automatically change the routting strategy. So let's say for example you were doing coding. Well, you can set the setting auto forward/coding and that will automatically use the best models for code generation because each model has different strengths and weaknesses and this way you can automatically switch between them based on the task you're doing. So if you're offline for example, you could also switch between those as well. And the main point is you have many options, probably more options than you realize when it comes to free models and there's many different ways to make the most of them. So for me personally, I like to use omni root. Our local engine is pretty good. We can use free local models inside Hermes agent as well. So for example, we have agents A1 which is a local model and we've plugged that into the Hermes. So if we have a gentic coding tasks, we can use Hermes locally for free as well. And if you want to get our full system with all of this set up plus token minimization strategies too so that not only can you use free models but you don't get rate limited then you can check out our agent operating system inside the AI prof link in the comments in description or go to the profarm.com. If you're wondering what was that GitHub before, it's called free LM API resources and this is the full list that you can get with all the details of each free model. So, hope to see you inside the air profitable boardroom. This is my community for learning, growing, and scaling with AI automation. Inside the community, I personally answer all the questions with video tutorials every day. Inside the classroom, you can get all of our new daily trainings that are free and you can learn from and we update this daily with new tutorials, new guides, etc. Inside the agent OS, you can get the full system here. You can see when it was last updated. So, we update it daily. And you can get the zip file to install with your agents plus a video tutorial. And every time something new and useful comes out like for example I was mentioning before MLX is now 90% faster with Gemma before you can see a full video tutorial and step-bystep guide on exactly how to use it. Inside the calendar you can jump a weekly coaching call, share your screen etc. And inside the map you can meet people in your local area who are building with AI agents like you. So feel free to get that link in the comments description or go to the appwarm.com. Guten Morgan. I haven't tried that, but we do have people inside the air proper who do that. Uh, so it's a great place to connect with people who are already doing things like that. And feel free to post inside the community if you want to meet some cool people doing that sort of stuff. All right, thanks for watching. I'll see you on the next one, peeps.