Best FREE GitHub Repos for AI

Today I'm going to show you the best GitHub repos for AI and these are trending new repositories. If you don't know what these are, don't worry, I didn't even know what they were until a year or two ago when I started getting into AI. But basically what these are is these are open source projects, right? These are projects basically like apps you can download that are shared publicly and so anyone can use them, anyone can build on them.

So for example, if you have a look at Hermes Agent, this is by far, this is absolute fire, right? This is one of the best new GitHub repos that came out. It only came out in February and it's pretty much changed the world, I would say, at this point. It's a competitor to OpenClaw and this is a really powerful AI agent that you can use for AI and you can use it in Telegram, you can use it in your terminal, etc, right?

And it's a free repository. So basically these are all different AI projects that you can use and install and get access to for free and you can start using them today. So I'm going to walk you through some of the best ones, some of the most trending ones and talk you through exactly what they do, what they are, how to install them, etc. How to get access to them.

So let's kick this off and we'll kick it off with Hermes Agent, which is one of the fastest growing repos I've seen for a long time. Bear in mind, again, only launched in the end of February and you can see it's already at 72,800 stars on GitHub, right? So Hermes AI Agent is a powerful AI agent that you can get access to right now. Now this is quite different from OpenClaw in the fact that it self-learns, it self-improves, right?

So it has a positive feedback loop where basically every mistake it makes, it writes it down. Now there's some other stuff as well. So for example, you can create multiple profiles with your AI agents and then you can have them chat to each other. I'll show you exactly what it means.

So for example, if we have a look at Hermes over here, this is Hermes inside my Telegram, you can see that these Hermes Agents, I've added two agents inside the group and they've sort of like started talking to each other, right? Pretty crazy. And so what we're doing here is we are running AI agents inside my Telegram that I can access from anywhere in the world and they can do scheduled tasks. So for example, each day they can organize and do competitive research for me, they can create videos for me, they can take analytics for me, as you can see right here, right?

They can search across the web, they can even post like tweets for me and stuff like that, right? And so Hermes is a really powerful AI agent, kind of like OpenClaw, but I would genuinely say it's better when I've tested them both simply because it's less buggy. Now if you want to get started with installing this, just follow the quick install rules here, right? And once you've done that, you can go into your terminal like this and if we open up a new window, we type in Hermes, it's as simple as this.

It's going to start up Hermes inside our terminal right here, right? And so it's a really powerful self-improving AI agent that you can get access to right now. Now if we have a look on OpenRouter, I'll show you how big this actually is, right? So if we have a look at the apps, and OpenRouter is like, you know, it's one of the biggest platforms in the world for AI.

If we have a look at the most popular apps, you've got Claude Code, which, you know, most people watching this channel will know of. OpenClaw, which has been absolutely massive. But look at that, Hermes Agent is in the top four apps and this, again, only launched end of February. So this is one of the trendiest and one of the most powerful AI agents I've seen.

It's grown so quickly and I think that basically just shows you that people absolutely love it, right? So this is one of the best AI tools, I would say, because it's a free AI agent. You know, you can see here that I've actually got Hermes running as well and I've got Manus running, right? Manus is another AI agent that could create content for me, it can post to social media, etc.

But the problem and the trouble with Manus is, like, it's pretty expensive to run, right? Whereas, for example, if we use Hermes, that's free to run, right? And we can even run it locally with local models like Jemma 4 and that sort of thing. So if you're interested in this sort of stuff and you're technical and you like to, you know, sort of play and tinker with different things, this self-improvement agent by Hermes and Noose Research is great.

Also, one thing to note here is it is created by Noose Research. Noose Research is one of the biggest AI labs in the world, right? They create their own models as well as their own AI agents and they're very well respected and so it's definitely worth checking that out. Alright, so that's number one.

Let's move on to the next one. So there's something else called Andrej Karpathy Skills, right? This is another GitHub repo that's become very, very popular recently. So if you're wondering, okay, what is this?

Basically, Andrej Karpathy was the head of AI at Tesla and one of the founders of AI, right? And this is one of the best repos that I've seen come out recently. What I like about Andrej Karpathy is, like, he releases really simple stuff and he has these interesting observations on AI, right? As being one of the founders, I think he's just got so much social proof that he has a very powerful voice in AI.

You see this tweet actually got 7.6 million views already that he posted recently, right? And so basically he had a few ideas in terms of, like, the mistakes and the problems with AI and, you know, the limitations of it. And so someone actually created a GitHub repo that manages that, right? That fixes those problems.

So you can see, for example, these give Cloud Code guidelines on exactly how to fix the problems that Andrej Karpathy has found, right? So I'll give you some examples. You can see here that he complains that AI, like, is overcomplicated stuff. It often makes wrong assumptions, which it does.

And also sometimes it edits everything instead of just editing one thing, instead of saying laser focus, right? So you'll say, for example, like, improve this website, but Andrej Karpathy, sorry, Cloud Code will see that and edit everything, not just, like, one single part of the page that needs to be improved. Now the solution is four different principles. And basically these four principles come in this Cloud MD file.

And so when Cloud starts being used, it follows these principles. Now, if you want to install it, you just go over to Cloud like this. So you can go inside your terminal here and we'll open up a new tab. We'll type in Cloud to run Cloud Code, right?

We'll pull this up side by side so you can see what I'm talking about. And then inside here, for example, we can say, okay, install this, and then we'll just get it to install this Cloud MD file. There's probably a faster way to do it, but when I'm installing GitHub repos, I really like to use Cloud Code because it just figures everything out for me. And I think if you're non-technical, if you're not a technical person, this is one of the best ways to quickly install GitHub repos, right?

So now this is running, as you can see. And basically there's four principles inside this Cloud MD file. Now, basically, to simplify it as much as possible, Cloud will use this set of rules every time you code inside it, or every time you use it for any sort of help, right? And so it's going to follow four principles to make it much smarter and much more powerful.

Method number one, principle number one, is think before coding, right? So any sort of wrong assumption, it's going to handle those pretty quickly. Now, you can see the full breakdown here. So think before coding, this principle right here is like, it won't assume, it's not going to hide something, it's going to surface any problems, and it's going to state any assumptions explicitly every time it's being used, right?

You can see here, number two, it's going to present multiple interpretations, not just pick silently when there's two choices. It's going to push back when warranted, and it's going to stop when it's confused. So that is principle number one that just makes it a lot better. Principle number two is simplicity first, right?

So minimum code that solves a problem, nothing speculative. So it's going to combat the tendency toward over-engineering. Basically what this means is like AI typically over-complex things, and it does things like in a very difficult way rather than just being very simple. And you know the reason why it does this is because AI has been trained on human data, right?

The AI only hallucinates and does things wrong because it's been trained on humans and what we do, right? So it's our fault really. But number two is really simplicity first, right? So it's not going to add any features beyond what was asked.

It's not going to do any extractions for single use code. It's going to have, you know, it's going to keep the code really lean when it's being, when it's writing code for a page, right? So that's number two. And basically the test here is like, would a senior engineer say this is over-complicated?

If yes, you need to simplify the project even more. Now rule number three inside this Claude MD file, and it literally is very simple, it's just a set of rules, but it does make everything better, right? It's kind of like a meta improvement where one single improvement changes everything from that point. And that's basically what we're doing with Claude code when we install this.

And bear in mind that you don't have to just use it inside Claude code. You could take this MD file and give it to OpenClaude as a skill. You could give it to Hermes as a skill, any AI agent that you use, right? So surgical changes.

Touch only what you must. Clean up only your own mess. So basically when it's doing this is like, when it's editing existing code, it's not going to refactor things that aren't broken. It's going to match the existing style, right?

And it's just going to, the test here is like every change line, anything that you change inside the code, so anytime Claude makes changes, trace it directly to the user's request. And then number four is goal-driven execution. So define the success criteria, loop until verified, right? Now this is an interesting one because basically what it's doing here is when humans give an AI agent a task, quite often we're pretty bad at telling it what done and finished looks like, right?

When we delegate anything to anyone, right? Even to humans, we don't tell people like, okay, this is what finished looks like. And we should really, right? If we want to get the best outputs.

And so what we can do here is it focuses on goal-driven stuff and it will clarify it until we finally confirm the goal. And it wants strong success criteria so that the LLM can just loop independently until it hits that goal, right? And so essentially, if I say to Claude, you know, write me a blog post, I don't really tell it when it's finished. If I say, write me a blog post with, you know, simplified third grade language, SEO optimization on the page, and also make sure it's 2000 words, right?

That's a lot more clear in terms of a goal and in terms of what finished looks like. And so it'll loop around until it's finally done that. And so it uses all those four tasks inside this GitHub repo to improve. So that is repo number two.

Now there's actually so much more to this. So for example, there's loads of other ones. I want to have a look here and see what else we've got on the list. SEO machine is pretty interesting.

If you're into SEO or AI SEO, basically this turns Claude Code workspace into a way to create SEO optimized blog content for any business. Really cool skill. I've seen it absolutely blow up recently. And basically what this allows you to do is give a ton of skills and a ton of different tasks to Claude Code to train it up on how to do SEO, right?

And one of the interesting things about this is like, there's a lot of tasks in SEO that are super boring, super time consuming, super draining. And so you can use Claude Code to handle those, right? And it can run the Python to actually do this. So for example, it's got Google Analytics and Search Console integrations, data for SEO, API clients, machine learning, et cetera.

And you can just run it inside Claude Code. So it just makes Claude Code way more powerful, actually doing SEO and helping you rank your website. Now you can see here with the trending repos, we can search by this week, this month, this day, et cetera. And things change a lot, right?

Now, one thing that I have seen that looks really cool is Everything Claude Code. And basically what this is, is a way of just making Claude Code a hundred times better, right? So Everything Claude Code, it was created by an Anthropic Hackathon winner, right? It's already got 140,000 stars on GitHub.

And basically Everything Claude Code, what it does is it optimizes your AI agent harnesses, right? So Claude is an AI agent harness. It contains an AI agent, which is Claude, right? And the reason that Claude Code is so good is because it just makes Claude's API 10 times better.

So using this GitHub repo, this makes Claude way more powerful because it gives it all these other skills and all these other ways of running and optimizing Claude Code to get the most out of it, right? Now, if you want to install this, you just take the GitHub repo like so, and you would just say, okay, install this, right? And that's basically how it works. Now, this includes like skills, instincts, memory optimization, continuous learning, security scanning, research-first development, and also you get these production-ready agents, skills, hooks, and rules, right?

So it's got tons of different angles to there. It's not just like an AI agent or something like that. And also you don't just have to use it inside Claude Code. You could use this inside Cursor, inside Codex, inside OpenCode, inside Gemini, right?

Any sort of AI agent harness, aka any sort of system that controls an AI, you can plug this into and install it. And it's free, right? All these GitHub repos are free. Sometimes you need an API to run them, but you can always get a free local API if you want to run something for cheaper, right?

And so what this helps you with, as you can see right here, is it has like a bunch of guides, right? So you can actually learn, for example, how to optimize for tokens, right? Which means how do you reduce your costs on Claude. If you ever hit like a token limit on Claude, you can optimize it and reduce it by using token optimization.

There's also memory persistence. So this allows your Claude to remember across multiple sessions. Continuous learning, so you can learn how to just make sure that Claude is always improving. And sub-agent orchestration, right?

And so inside this GitHub repo, you get all these different skills, all these different things that just make Claude Code 100 times better, right? And you can do that by just installing the GitHub repo. So these are some of my favorite GitHub repos right now. This is a very interesting one, MyraFish.

This basically allows you to run predictions. So you can say, okay, like I'm thinking about doing this, or I'm thinking about doing that, etc. Run a swarm of agents, have them interact with each other, and then MyraFish will actually run that together. So these AI agents run together and sort of debate between each other on the topic or the decision you're trying to make.

And it runs for such a long time that eventually you get a consensus in terms of, like, what do you think the output would be? Then you've got superpowers as well. This is pretty cool. So superpowers is an agentic skills framework that allows you to just improve the skills of your Claude Code.

Now, each of these skills, basically what it is, it's kind of like, think of it like the app store, right? If you download an app for your phone, it already has that capability to do it, but you need to train your phone on how to do it, right? How do you do that? You download an app, right?

So these are all kind of like mini apps. These skills are like mini apps that you can use with Claude Code. So for example, he's got like a brainstorming skill here, and you can check the skill MD file and just check through and see, okay, what skills can it do? What can it do?

What are the key principles of it, right? How is it structured? What actually is this about? And it's all written in plain English, kind of like a blog post.

It's nicely formatted so that it tells the agent exactly what to do. But also it means that you can check the skill before you install it and see like, okay, should I add that or not, right? Is that going to be useful or not? Or do I install this or do I create my own version that's customized for me, right?

That's a better way to do it. So either way, all these different skills are great. So Superpowers is the final one. Now, if you want to get any of these, you can get them for free on GitHub, all these amazing open source projects.

Honestly, there's new ones coming out all the time. I didn't even get time to mention OpenCore today, but you know, there's just so many different GitHub repos you can get for AI, and you can install all these into your preferred AI agent. If you want to learn more about this stuff, feel free to check out the AI Profit Boarding. This is my AI community where you can learn, grow and scale with AI automation.

You'll learn how to save time. You'll be able to ask any questions inside the community. You can jump on live coaching calls each week where we go deep on this sort of stuff. And also inside the map, you can connect with people in your local area who are doing similar things to you, right?

So if we go inside the map here, you can see that you can meet people in your city, DM them, be up with them. And these are people who are deep on core, deep on open source projects, etc. And then inside the classroom here as well, you can get all my best trainings on this sort of stuff with video tutorials and step-by-step guides. Pretty much everything that I've shown you today, you can get a step-by-step guide on as a video inside the AI Profit Boarding.

So feel free to get that link in the comments description or just go to the AIProfitBoarding.com. Thanks for watching.