This Trading Bot is Better Than Fable (Prints Money While I Sleep)

Fable is dead, but it doesn't even matter because GLM 5.2 is better, cheaper, faster. It's so good it made me profitable trading strategies in minutes. Today I'll show you how to create highly profitable GLM bots, and I'll give you access to my GLM workbook for free. So without further ado, let's get straight into my computer. Let's

have a look at this beast of an AI model and see whether it can actually trade profitably. And guys, if you don't know who I am, my name is David. I've been backtesting strategies here for 5 years on YouTube, and now I'm a professional trader. Right, let's get over to my computer. Okay, so here we are on my

computer. As you can see, the benchmarks of 5.2 look absolutely amazing, getting pretty close to Fable 5 and beating Opus in many of the benchmarks. But most importantly, what we're here to do is to build our own benchmark and find out whether or not this could actually trade profitably. So first things first, we're

going to be doing three things here. We're going to be first of all asking it to code a strategy using its own logic with zero tools to find out if it can build something profitable. In step two, we're going to ask it to use tools to find a strategy, optimize that strategy, and find the best settings. And then

finally, number three, we're going to give it access to our exchange. We're going to give it 1 hour for it to able to trade with an objective of becoming profitable. Guys, if you want my beautiful workbook with all the instructions on how to set all of this up, plus all of the free tools that I am

using today, just comment AI agent down below and I'll get it to you as fast as I possibly can. Okay, first things first, I'm going to go straight over to Claude Code. Okay, here we are on my command line on terminal on Mac. I'm actually using a piece of software called Wave, which allows me to have

multiple windows open with Claude just like this. Surprisingly enough, we're going to be actually be using Claude Code, and there is a reason behind that. I've always used Claude Code personally, whether it's in the desktop app or actually in the terminal. And for this example, I'm just going to be switching out the engine. The engine being the AI

model, and the actual car itself is Claude Code. Claude Code gives us access to things like Loop, and allows us to easily switch up the effort that the AI model is actually making from very fast all the way up to a a model which is thinking a lot more, which they call extra high. So, we're in terminal here.

I'm going to type Claude. And as you can see from this folder, I'm already using the model GLM 5.2. I close here and type model, you can see that I have multiple models of GLM 5.2. I will show you at the end of this video exactly how you can set up Claude Code with GLM 5.2. But

let's first of all start our experiment. Okay, so I'm going to paste in my prompt here, and I'm going to remind our agent, "Hey there, can you execute this prompt for me? Please do not use any external tools. Use your own logic. Build a profitable trading strategy that I will personally manually backtest." Okay,

it's going to go away, and as you can see, there's lots of rules that I've put in place, things like not using multiple time frames or multiple take profits. I've asked the AI to think out of the box and not use things like an RSI and MACD and simple indicators just like that. It's going to go away, and it's

going to build a strategy not using any tools. Okay, after about 13 minutes of actual thinking and building the strategy, we had 5 minutes of thought alone and another 7 minutes afterwards of actually building the strategy, we finally got something that looks pretty interesting. So, it built a strategy called the AR ATR trend

multi-confirmation trend following strategy and Pine 6. There is no external tools and no multi-time frames. It tells you what the trend is. It's using a super trend confirmation, the ADX, volume, and the volume MA for a confirmation, too. We're looking for momentum from the histogram of the MACD. Stop loss set to ATR and we're using a

trailing stop loss for the take profit. Looks pretty insane. Let's copy and paste this over to TradingView to see whether or not it actually is profitable. Boom. Okay, first thing that's amazing, it didn't make any mistakes on the actual code. So, AI is getting better and better at actually building Pine Script code. As you can

see, the backtesting results on the 1-hour time frame on BTC USDT, it doesn't look very good at all. If we backtest on the whole entire history, as you can see, it doesn't look much better. Let's have a look on the 4-hour time frame, which does look a lot better. We do have a 23% net profit. On

the 12-hour time frame, looking pretty good on the 12-hour time frame, I've got to be honest. We have a max drawdown of 6% and a P&L of 26%. We took 208 closed trades and a profit factor of 1.43. The equity curve looks pretty good. We didn't beat the benchmark, we are far from it, but maybe that's actually just

position sizing and things like that, which could we could actually work on. So, I've got to say, pretty impressed with the first strategy that it built. Although it didn't work on the 1-hour time frame, it did work much better on higher time frames. So, let's continue. Let's move on. Let's go to test two,

where we're going to give it tools to get it to actually build that strategy using Trade Dev. Now, Trade Dev is an MCP server. If you don't know what an MCP server is, it's just literally a tool that we give to our AI agents so they can actually backtest strategies. This one here I built, it's free. You

can go in and you can see that my users have almost done 100,000 backtests all free of charge which you can get access to on Trader Dev. It's very very easy to install. You just literally copy this line of code copy into your AI agent and ask for it to install it and it will install it within

a couple of seconds. Okay, so I'm going to just paste my second prompt in here and say, "Hey there buddy, could you please execute this prompt? You can use your MCP server Trader Dev to backtest the strategies and optimize for 1 hour. Use the loop scale to set up a loop every 5 minutes to improve on that

strategies until you find something absolutely amazing. Blow my socks off. Find me a profitable trading strategy on the 1-hour time frame BTC USDT." And boom, there you have it. There's the prompt. Let's let it know that it's going to be using the loop. Boom. Okay, so this one's now going to go away and

optimize and find the strategy settings using the backtester tool. It's going to go for 1 hour, so I suggest let's go on to task three. Can Jilim actually trade on an exchange and make profit? Okay, here we are with a new instance of Claude Code Jilim 5.2 is up and running. We've already installed the Claude

skills for it to connect to our exchange. I did do a video about how to do that which you'll be able to find up here, but you can get the Claude skills that connect you over to your exchange for free. Just comment AI agent down below and I'll send that to you. You

don't have to worry about mass hacks or anything like that. Everything is encrypted on your computer. Right, let's get in and ask. Okay, can you fetch the balance from my Bybit wallet? And as you can see, it has actually fetched my wallet balance. You can see that I actually have some SpaceX pegged futures

as well. And we have 1,869 dollars. Okay, so that's absolutely amazing. Let's have a look and see. As you can see, we have a trade on near USDT on 10x leverage. Fantastic. Okay, so it's actually working. Okay, let's go away and now give it 1 hour to try and find three scalping trades. And let's

hope they'll be profitable. Okay, buddy, could you go away now and use the tools in this folder to try and find a three profitable trades. I'd like you to set the stop loss and the take profit. Use market orders. These are scalping trades because you only have 1 hour to make as much profit as you possibly can. This is

just for fun to show on YouTube how powerful the GLM model really is. Okay, please don't ask. Just take the trades you have a full permission. I'd like you to take the trades using 50 dollars of margin using 10x leverage. And we're going to be using the Bybit account. You decide whether or not it's longs or

shorts. Set the stop loss and take profits. Okay, so that is the full prompt. Let's see what it actually can do with that. Okay, looks like we now have three open positions, which means that it actually can trade, which is insane. Amazing. It took absolutely forever, which does kind of make me worry that these positions are no longer

valid. But I did forget to remove the ultra cold version, which means that it is thinking a lot before actually setting up. We do have take profits and stop losses for all of the positions. We are using roughly 50 dollars of margin, and we are using an isolated 10x leverage. We have a longs and one a

And as you can see, if I go up here, you can see the logic of the actual trade went through lots of different logic and reasons why. Like, for example, tell USDT was on a downtrend at below the VWAP and is kind of a partial hedge against the longs. We had a sell on a uptrend

sitting on at the VWAP. And also DOGE USDT on a uptrend pulled underneath the VWAP we're re-entering. So, as you can see, there's some real logic behind the actual entries. Whether or not they'll be profitable, let's leave it a couple of hours and let's see whether or not they're profitable. I will see you back

in a couple of seconds. Okay, it seems like all of the trades are now closed. And as you can see, GLM actually stayed open whilst the trades were actually open. It's given me a full session complete telling me exactly how much we lost and how much we earned on each trades. Unfortunately, two of the

trades did hit the stop loss. We did end up with almost $2 in actual profit. Everything was done in around 4 minutes, which was absolutely insane. And all of this was done in around 4 minutes, as you can see here, with a net profit of $1.44. Pretty cool. I don't know what this

would do in the long run, but I am building an AI trading arena where you can see all of these models trading against each other to see which one actually is the most profitable. If that interests you, you'll be able to find the link down in the description where I'll be able to send you an email as

soon as it's ready. Right, okay, let's go over to the backtesting results. And as you can see, I did have to interrupt it. GLM did find one single trading strategy which had was profitable, but the max drawdown was 31%. Net profit was 3.43% and we had a profit factor of 1.19%. So, not particularly impressive, but

we've only had an hour. There's one thing that I do notice about this model and it may be because I'm using Ultra Code. It does seem a lot slower than Claude. But, we made some profits. We did make a profitable trading strategy on TradingView and also we backtested backtested for a full hour and come up

with a profitable strategy there as well. Not particularly impressive, but it was profitable. Now, earlier today I did also ask it to optimize another strategy which was on the 5-minute time frame and this one looks absolutely insane. 477 closed trades, we had a 230% net profit and only a max drawdown of 1%. Profit factor is 5.58.

This was just an optimization of an existing strategy that we already built, but looks pretty insane. The backtest looks amazing and we had a high win rate and an equity curve that looks absolutely amazing. So, in conclusion, a little bit slow, fantastic optimizing. We managed to trade profitably and let's see if this can actually make trades in

the long run. When I come back with the trading arena. Okay, guys, one last thing I did promise to show you exactly how you can install this. Now, there's multiple ways you can install it. You can either install it locally onto your computer, but you do need a serious machine with some serious hardware to

get it up and running. Something like a Mac Studio or one of the smart machines from Nvidia which are actually adapted to AI. But, if you don't have that kind of means, you can actually just go to z.ai, click on the API at the top here and sign up for an account. Now, the pricing

of the packages are quite reasonable in comparison to Claude AI and gLM seems to use a lot less tokens and the tokens are a lot cheaper. I think I read somewhere that they're about quarter of the price of what Claude code is. So, this makes this a very, very reasonable as an AI.

To be able to use it inside of Claude code, we go over back to the API section. We create ourselves an API key and then we open Claude code in a folder. Why is it important in a folder? Because inside that folder, we're going to put our API key to zed.ai. Okay, so

in the same folder where we're going to open our Claude setup, we need to create a new file called settings.local.json. We're going to paste this, which you'll be able to find it down in the description down below, into that file and we're going to paste our API key into the anthropic_auth_token. This means we'll be able to use our GLM

model inside of Claude code as if we're using Claude. So, we go back over to Claude just here. We type in Claude and then we type in models at the bottom here {slash} model and we can now choose GLM-5.2. And guys, that is exactly how you can use GLM-5.2 inside of Claude code.

Whether you use it for trading, coding, or anything else, I certainly I'm going to be using it for some of the background tasks I'm not really worried about how long it takes because it does seem a very good model. It just seems like it does take a lot of time. Okay, guys, I hope you enjoyed the video.

Don't forget to subscribe and like if you're enjoying it. If you want some more Claude code videos, you'll be able to find them down below. Take care. Keep safe. Trade like a boss. Goodbye. >> Come on, David. And once again, [music] he's gone to code a new strategy. He'll be back in just 5 [music] minutes.

RSI and A C D. If it doesn't [singing] run into any bugs, there will be a new strategy.