Transcription
The holy grail for AI trading agents is having an agent that's able to learn from its mistakes and make improvements on the strategy or with the vision of becoming more profitable. The thing is is that most AIs that you interact with, they are very simple. You give a prompt, they give an output. But what we're going to do today is use an extremely powerful AI that automatically learns from all the engagements that you have with it. And I want to see if I can apply that self-learning behavior to a trading strategy. So instead of it being prompt outcome, instead we're going to produce a prompt that creates a strategy that creates an outcome it can learn from and therefore creates a new prompt to build into the strategy. Again, what we have is a self-improving trading agent. You're going to have it running 24/7. And I've even made it so you can just simply copy and paste one single prompt, put it into your AI, and it will set this whole thing up for you. And the AI itself that we're using today is completely free. It does this self-eing process, this self-improvement process for free. So, right after you subscribe, let's get into it.
Now, I've been doing this a little while. And I know that whenever you want to create a self-arning or a self-improving process with Claude, for example, you have to describe how it's supposed to improve itself. And so, it's it can get a little bit laborious. It's very boring, and it's very frustrating because the AI doesn't quite understand it. However, there's a new tool that has come out over the last couple of months that's been really lingering in the background and today I've just seen the massive use case for it in trading and it's called Hermes agent. You might have heard of like OpenClaw, this fully autonomous thing that's just took the world by storm. Well, Hermes in the background is being touted as even better than OpenClaw because of this self-learning process. But Hermes agent on its own isn't going to do all the work. We actually have to put some smarts into this. And this is what I do. I architect agents to do exactly the thing I want them to do. And so there were four criteria that I came up with with regards to what makes a good trading agent. And I came up with these four. So number one, it has to be accurate. Number two, it has to be reliable. Number three, it needs to have a very well-defined goal. And number four, it needs to be self-improving. It needs to learn from its mistakes.
Now, let's first talk about the accuracy element because when I say accurate, I mean, is the data that's coming in accurate in the first place? Are we able to take that information in reliably and consistently over time? And when we feed that information to Hermes agent, is the information actually accurate? Over the last couple of weeks, I tested every single AI that is in existence. And I tested them on their ability to do trading. And one of the most shocking results from that is the inaccuracy in the data. They're all supposed to be pulling information from the same place, but some AIs just don't do it properly. And so there is inherently an accuracy issue. So we need to make sure that that is resolved, and we will resolve it in this build. So when we're trying to get accurate data through, we need to make sure that the API connections are very strong and reliable. We also need to make sure that if we're pulling in information from news feeds, for example, that that information is also accurate because the AI can sometimes interpret text in different ways. If you give the same article to multiple different agents, they might have different things to say and different conclusions. So, we need to make sure that there are rules in place so that the conclusions are accurate and ideally accurate and objective.
And that brings me on to the reliability section. We need this agent to be reliable. So, what do I mean by reliable? I would say that reliable is it's always operating 24/7 and even if our computer goes down or turns off or closes and it's still executing on the system that we're going to build today. We have also solved that issue in the oneshot prompt. Now, it's incredibly important that we move on to number three, and that is that we have to make sure that the agent has a well-defined goal. So, let's talk about goals a little bit deeper because it will make all the sense in the world to you in just a second. So, we need to define in terms of a goal, obviously, there is a destination. Most people, I'd say 90% of people right now, if you're creating a trading strategy, don't have a destination, don't have a definition of what achieving the goal actually looks like. And so in light of being thorough and actually accurate and actually having a good agent. So we need to define what is success and what is failure. And it might sound a bit abstract but in fact we need this information. What is success in the strategy? Is success making $10 a month? Is success making a million dollars a month? Obviously you have to operate in the in the realms of what's possible. An example of something that would be impossible would be like saying I want to make a million dollars a month and here's $10 to start with. So what is success? And the more details you can give here, the better. Like if you know anything about sharp scores, which is essentially a a score that relates to the profitability of a a trading strategy, you would might want to put a specific sharp score into the agent as a goal, right? We want to work towards this goal. Cuz if you think about it, this agent's going to be doing a thing, getting feedback, and then improving the thing to do it again. And it's going to do that over and over and over again until that goal is achieved. And so we need to define the goal, right? But we also need to determine what is failure. What does failure look like? And so what the agent will be able to do in the end is almost like look look at where it is in its current results and say okay anything in this direction is towards the goal and this is good and anything in the wrong direction away from the goal closer to failure is bad.
And it's with that goal in mind that number four comes in and that is that it needs to be self-improving. So it needs to be able to assemble and organize information properly. It needs to learn from the outcomes. It needs to analyze the outcomes. Were they towards the goal or away from the goal? It then needs to form its own hypothesis about why the result was the way it was based on the information it had. And then it needs to make a second hypothesis about what it should do next. So then it should take that information and that learning apply it to a updated strategy. And this updated strategy, I think, should follow the scientific model, which if you don't know what the scientific method is, it's essentially changing only one variable and then seeing the outcome. Because if you changed a load of variables and you went more profitable, you wouldn't know which variable was responsible for that trade going well. And so you only change one variable at a time and you run a series of tests. Every time you get one better, that is now the new baseline. And then you make iterations on that new baseline. And it needs to do this inherently. And so all four of those things make up what I think is a good agent.
So now's the time that we're going to start creating this. I'm going to demo it for you from start to finish. The setup of this thing, the prompt, everything is completely free for you to use. I'm going to give you everything that you need to copy and paste and get this agent up and running with Hermes. Okay. So, as always, every single prompt is freely available for you to copy and paste. And I hold them all in 01 systems. It's my own free community that you can join right now. The link is in the top line of this of the description. And anytime I post any prompts in any future videos, the links and the prompts will all be in 01 systems. You'll come here to start with. You'll click classroom at the top. Then we're going to click this big YouTube button. This is for all the YouTube video prompts. And you'll come across something titled something similar to this. Self-improving trading agent. Hermes self-improving trading agent. The video will also be in here cuz you can see this is how I post it when it's live. And we'll open that up. And we're going to take this beautiful oneshot prompt. It is so nice. And by the way, all of the oneshot prompts that I give in my videos, they improve over time as well because we get feedback. People have certain issues and then we improve them. So the version that you're downloading right now or that you get in 01 systems will be the most up to date and the best one we've had so far. And it's only getting better. It's so cool.
Okay. So, what we're going to do is we're going to come over to our terminal. And this is me in my terminal. I'm just going to increase the size so you can see it. We're going to start a new session, which I do with dangerously skip permissions because I'm an absolute savage. Okay. And then here we are. We're going to we're in Claude and we're just going to paste in our oneshot prompt. So get ready because the journey begins for your self-arning, self-improving agent that's going to run on Hermes. So as we go through this process, we're going to do a series of phases which you'll see on the screen right now. So phase one was an environment check. What this does, it's really cool, is it makes sure uh it knows which system you're on. Are you on a Mac or on a Windows? And depending on which one you choose, then it will take you on a different journey because there's different instructions for both. So it said, okay, we can see that you're on a Mac and you've got no.js. JS installed on clawed code. Great. Step two of seven in phase two, which is defining the strategy. We're going to build your trading strategy now. Specifically, what success and failures look like. The the agent uses this file to score every trade. It's not just vibes, it's just numbers. So, we have to now decide which asset are we going to be trading. Now, this is the moment where if you already have a strategy, you come down to number four and you actually you actually would say something along the lines of this. Hey, I've actually already got a strategy and it's called the Wacko Alpha strategy. So, could you look for that in my computer and, you know, use that as part of this system? Or alternatively, you could say, I don't have a strategy. Can you just make me a basic one? And it will make you a basic one. Uh, like a basic solid one that everyone kind of starts with and then you can let the agent improve it over time rather than you. Alternatively, you can build the strategy in this system. The onboarding agent will work with you to create a strategy, too. So, you could choose Salana or USD or Ethereum or Bitcoin or any asset. Uh, but for now, I'm just going to say number four. I've actually already got a strategy. It's going to call in to my information about that strategy that I already have, and it's going to build out my documentation based on that, which I think is so cool. So, let's let this work for a little while. We're going to go onto phase three after it's found my Wacko Alpha strategy. You can see it actually has found it right here. And what's wonderful is I've created this strategy already and it's got over a million and a half data points that it's analyzed. I've been letting it run for like six to eight weeks just learning from the information and I did that all manually like instructing how to learn this stuff. But the Hermes agent will just learn it itself. So it says what I'm seeing on the disc got wacko alpha the dwal momentum and yield strategy. Um how do I how do you want me to incorporate wacko alpha into this Hermes deploy? Yeah, let's actually point it at this strategy. Like, let's actually This is real money that's being traded, by the way. So, maybe this is maybe a bit of a mistake, but by the way, you can see the progress of my £50,000 to £500,000 in a year challenge that I'm doing. I'll call it the the 10X challenge. You can see that the dashboard is linked below. You can see my progress. The frustration is for me is that I haven't been able to put as much money in as I wanted to. I haven't had the dips in the market that I wanted to make my purchases. So, that's kind of been a little bit difficult.
Okay, so now it's actually pulled out the goals and said the maximum return 30 days is this much 10x in 6 months uh 40. It's defining all my stuff. My my minimum sharp score, my max draw down, my failure below reflection every certain amount of days. It's got all of this built in. So that's wonderful. And it says, do you want to confirm the Hermes and Wacko Alpha setup? Yes. So I'm going to have lock in as proposed. That's what I'm going to go for cuz this is actually real money. But I want you to know that I trust this system and the way that it learns uh sufficiently to put my real money on the line. That's what we're doing here. And let's move on to the next phase.
Okay. So phase three is now scaffolding the Hermes side state. So scaffolding all the folders and files to be properly analyzed by Hermes when we actually come to install Hermes. Okay. Now we're on phase four which has skipped actually because the Wacko alpha is already deployed. If you hadn't got a strategy already, it would start to deploy that strategy. uh to make it live. You might have to connect in like APIs or whatever to make it trade for you. But I've got a video on how to actually make things trade. It's like clawed code with trading view that actually trades. You can also find that in 01 systems, by the way. The whole prompt for that is there. So, I'm having a little bit of an issue logging into railway, which is going to be the place where we host this 24/7. So, it can run regardless of whether the computer's on or not. It's saying it can't run interactive login from inside this session. Please run this in the prompt yourself. So, all I'm going to do is going to come over here in my cursor, split this terminal so I can start a new terminal session. And I'm just pasting in this. So, I'm going to paste that in. It should then open up Railway for me to get me to log in. And it's a success. So, I can close the page. Boom. That's done. I can now close that. And I'm now logged in. So, I can say done, continuing. I love these oneshot prompts because I built them so that it opens up these browsers for you. So, if you don't have a Railway account, by the way, what would have happened just then is that it would have opened up Railway and you just make an account, right? And then you come back and say, "Hey, I've just made an account." And it's free for so much usage. Like, I've only just now started having to pay for Railway because I and I've cuz I've got like 50 projects on there running 24/7, right? So, it's now using the CLI, it's called, to integrate with Railway. So anytime you publish a new strategy or you make a change, it will update on the 24/7 server and just that's it. Just kind of works like that. So that's what's great about Railway cuz it works in this way with your terminal. Any project that you're doing is just kind of updating. It's now just seen uh 24 gain trades and 22 loss trades and it's converting those into a Hermes readable ledger. So it's now converting everything to be perfectly primed and ready for Hermes to take a look at it. Uh something else has just happened here. Oh yes, these new documents have just uh come up. Sorry about my desktop. Let's hide all the clutter. So my strategy document has now been populated. Um it's also o opening them in my cursor as well. That's really helpful. It's got my strategy. This is my strategy. The maximum amount of positions I'm going to hold is 12. My slippage tolerance is this. My gas reserve is this. The scorer weights everything. This is all pulled from my actual uh strategy. It's also defined my goals, right? My target uh return over 30 days is 4.7 which is 47% by the way. Um that's my target return for every 30 days. So it's going to be working until it achieves these goals. These are all the trades that have taken place as well. How cool is that? So it's just organizing all these files now so Hermes can have a look at it and and learn from it. So right now we're in the handoff to Hermes phase like just like that. So I hope you're if you're following along I think it's about time to subscribe, don't you?
Anyway, so it's basically looking at it's we're Oh my goodness, it already installed Hermes. It already did it. It already installed Hermes. That is insane. Actually, I didn't realize that would just happen so quickly. Okay, so Hermes has now been installed. I can now type in Hermes in this terminal or any other terminal. So, let's go over to the split terminal again. And I'm just going to type Hermes in here just to see if it is in fact I can't cuz I can't quite believe that it did install. It did. And now I fully have Hermes just up and running. That was just so quick. We could do a whole video on Hermes, by the way. It's unbelievable. But what it's done right now is it has outlined everything that it's doing. So now that we know Hermes is actually installed, we can come back to that later. So let's have a look at everything. So final confirmation, we have a self-improving trading agent which is deployed right now and adapted for my Wacko Alpha strategy which trade real money by the way. It's working on railway 24/7 and its strategy is using the bit tensor subnetss. We're looking for that return. We're looking for that as a max draw down minimum sharp of one. The brain is Hermes. It's going to be watching the live service and weekly cadence. Hermes owns the portfolio mechanics and score weights. And Cornelius, who's another agent of mine, owns the filter thresholds. The first cycle is read only and review only. It's going to flip the strategy uh YAML mode to live when it's ready. So, it's actually not trading yet. And when it's ready, Hermes will decide it's time to go, baby. And then we'll start making some money. Okay.
So, what happens from here? My strategy will keep firing every 30 minutes on railway, which it does already. It does a daily reshuffle and a 30 minute reshuffle. Cornelius, my other agent, is going to keep tuning the learned parameters JSON every week. That means he's like looking at all the data that comes in, those one and a half million data points that we have right now, and kind of learning from those. So, I bet Cornelius and Hermes now are just kind of together. Hermes reviews the trades weekly, but he has a 3-day offset from Cornelius. Interesting. The first Hermes cycle will produce a markdown review with no actual writing. Um, I will approve by setting mode. Um, so I can come here and just like approve the strategy from there. And a day after check-in, I can do any of the check-ins that I want using these commands. To go live, not today, is to edit the Hermes trading strategy. And Hermes will start writing on the next weekly cycle. Hermes is watching. Close this terminal. The agent is running. Kablam. Kaboom. There we go. What about that then? So cool.
Okay. So, I really want to hear your stories about this when you've installed. The best place to come and do that is to come into the classroom. There's another person that's joined who's actually taken on the 60-day challenge. Come into the YouTube video prompts. Take your prompt here. come over to the community once you've done it and tell us all about it, how cool it was and that that it works wonderfully the response of the last uh video regarding the uh Marov uh method strategy. People have absolutely you can see like no one has any issues with this thing. They loved it and they're using it and people are very excited about it. But let's see the response of this one. So we'll have a similar post like that on the community when you come in. Anyway, I want to thank you for being here. This is so much fun, right? um click like and if you can there's actually a capability of you to do something called a hype. So if you got this far in the video hype the video on a on the phone you can kind of like scroll across. If you click like the hype button will appear below. I think it really helps viewership. So give it a go. You can also hype on a computer but I don't know where it is. So anyway that's all. Uh have fun with it. Right.