Jev + Claude Opus 5.5 = The Most Powerful AI Trading Bot Ever

0:00Using Jev and Claude together for AI

0:01trading is insane. As you can see in

0:03front of me right now, I have a strategy

0:05that is running. This is high-frequency

0:07trading, so as you can see, there are

0:08live buys and live sells happening on

0:10the hyperliquid asset here, but you

0:12could choose any asset to run an algo

0:14strategy on. The brain that ideated the

0:16strategy is Claude, but Jev is actually

0:18handling execution. And the reason for

0:20this is that Jev costs a fraction of a

0:22cent every time it thinks. So, instead

0:24of needing to use inference on Claude,

0:26which costs a lot of money and takes

0:28more time, Jev is amazing high-frequency

0:30trading because it takes less time to

0:33make a decision, meaning you have less

0:35lag or latency. As you can see here, we

0:36have a very low latency, 400

0:38milliseconds, which is great for news

0:40trading, which I'll show you later in

0:41today's video. And it's also enabling a

0:43higher level of accuracy. So, in today's

0:45video, I'm going to break down exactly

0:47how you can start an AI trading bot

0:49using Claude plus Jev, exactly how it

0:51works, and I'm going to give you the

0:52full setup prompt. So, by the end of

0:54today, you are going to have a working

0:56trading bot as well. So, what is Jev?

0:58Well, Jev is not a traditional LLM like

1:00Claude or ChatGPT, so it doesn't

1:02generate any text. Instead, the creators

1:05of Jev refer to it as a system-one

1:07model, which are essentially a class of

1:09AI models built to make fast, structured

1:11decisions that software can use

1:13directly. So, a system-one model

1:14evaluates a state and returns typed

1:17answers and probabilities. So, every

1:19single time you present Jev with a

1:20choice, it places that choice on a

1:22scale, assigns a probability score, and

1:25if that probability score aligns with

1:28your preset instructions, it can trigger

1:30an action. AKA, in this case, a trade.

1:32And you can see right now, we have a

1:34bunch of trades firing. Now, for full

1:36transparency, this is running on a paper

1:37account right now, so it is a real

1:39strategy, but it's running on paper. I'm

1:41going to show you how you could actually

1:42run this on a live account because it's

1:44the exact same setup to actually make it

1:46live. Live meaning using real money and

1:49not just using a paper account balance.

1:51But you can see it's generated already

1:52$15 in profit since we started recording

1:55the video. So, Jev shouldn't replace an

1:57LLM. The real power of Jev is actually

1:59combining it with an LLM. Claude and

2:01ChatGPT, they're smart. They're great

2:03for inference. So, if you present them a

2:05problem, they'll find you a solution. I

2:06use Astra and Opus for designing all of

2:09my backtests, all of my strategies, for

2:11taking in the data, working out what

2:13works and what didn't. But, where I'm

2:14now using Jev in my trading bots is for

2:17handling the classification tasks at a

2:19lower latency and a lower cost, so I can

2:22execute trades quicker at a lower cost.

2:24So, this is why it becomes such a nice

2:26complement alongside an LLM, because I

2:28can use the LLM for open-ended reasoning

2:30and generation, and I can use Jev for

2:32fast structured decisions to actually

2:34execute trades. And for a high-frequency

2:36trading strategy, this is very powerful,

2:38because with high-frequency trading and

2:40news trading, you need very short

2:42response times, and you need to classify

2:44things very quickly. In terms of news

2:46trading specifically, it's all based on

2:48probabilities. If you're trading based

2:49on a headline that comes out, you need

2:51fast classification. You're talking in

2:53the milliseconds in terms of getting an

2:54edge. So, you can actually place a trade

2:56while there's still an edge on the

2:57table. So, essentially, you'll work with

2:59your LLM to develop the full strategy.

3:01So, pricing and sizing, risk limits,

3:03stop-loss, what exact strategy you want,

3:06and then you'll use Jev to take those

3:07variables and assign a probability to

3:09them, which then triggers a trade. And

3:11that's the engine which is powering this

3:12strategy in front of you. Now, if you're

3:14watching this video and you're like,

3:15"Oh, this is similar to Louis Jackson's

3:16video." It is. He actually inspired me,

3:19and going back one step before that, I

3:21follow Roan on X who comes up with

3:22amazing strategies, and he actually, I

3:24think, posted originally. So, I've taken

3:26inspiration from both of them, but I've

3:28added my own twist. I've made some

3:29improvements to Louis's engine, and I'm

3:32also going to show you another example

3:33in this video which neither of them

3:35showed, which is around news trading.

3:37And why this can be effective for

3:38actually setting up automated news

3:40trading. I've actually developed an

3:41engine which I'll show you today, which

3:42you can build yourself with a one-prompt

3:44setup guide for news trading. And I

3:46think that's where Jev's going to give a

3:47lot of people an edge. But, make sure

3:48you check out Roan if you're on X. Make

3:50sure you check out Louis if you're on

3:51YouTube. They're fantastic creators who

3:52are also deep in the algo trading space.

3:54And if you don't know much about me,

3:56I've been building trading bots since

3:572023. I think I was one of the first

3:59people on YouTube to put out videos

4:00about trading bots. I remember when I

4:02was doing strategies, backtesting Pine

4:04scripts on the old ChatGPT. We are

4:05progressing so fast now. So, I'm trying

4:07to keep you guys in the loop. So, make

4:08sure you subscribe because me and my

4:10team are constantly making tweaks and

4:12changes based on the new models that are

4:14coming out, which are much smarter, and

4:16based on the new tools coming out, which

4:17are making things much quicker. And I

4:19think there's just such a massive edge

4:20whether you're a trader or whether

4:21you're an investor implementing this AI

4:23technology into your workflows and

4:25systems. So, to set this up for

4:26yourself, all you need to do is join my

4:28free school community, and that's where

4:30I'm going to leave the one prompt setup

4:32guide. So, if you click on classroom,

4:33free market assets library, and then

4:35scroll down to the latest video. It

4:36should be down here depending on when

4:38you're watching it. You can then claim

4:39the .md file, which will be available

4:41once this video is uploaded. And that MD

4:43file is a starter prompt. So, it

4:45contains, and we'll keep this updated

4:47for weeks after this video comes out to

4:48make sure that it works for you guys.

4:50The full starter setup prompt that will

4:52build the Jev architecture for you. So,

4:54it will install Jev. You will need to do

4:55it on Vercel. But, my starter prompt

4:57will walk you through exactly what you

4:58need to do. So, you don't need to figure

5:00it out yourself. What it also contains

5:01is a spec for the dashboard, which I

5:03refined. I made some improvements on it,

5:05which is this strategy engine that

5:06you're seeing here where you can just

5:08plug in any strategy into the back end,

5:10which I'll get into in a second. You can

5:11do it on Claude. You can actually also

5:12do this on Codex. Drag and drop the

5:14starter prompt, once again, for free

5:16down below in my school community. Click

5:17enter, and then it will start up for

5:20you. And let me know in the comments how

5:22it works for you because we will be

5:23making tweaks as well over the coming

5:25weeks. The team will just update the

5:26prompt, and we'll just leave the latest

5:28version of the prompt in the school

5:29community at all times. Maybe at this

5:31stage in the video, you're still

5:32wondering, "Oh, Miles, is Jev just a

5:34gimmick? We hear about all these new AI

5:36tools all the time, and everyone always

5:37hypes up the latest model releases."

5:39But, I'm going to tell you right now,

5:40this genuinely is a game-changer for

5:42both trading and business workflows. So,

5:44I'm going to put out a business video on

5:45AI Edge about how it's actually changed

5:47how I run my business because it's the

5:49first time that AI can actually make

5:50decisions cheaply and quickly. And those

5:52are prerequisites that are so important

5:54for trading execution. So, let me

5:55explain to you very simply how this is

5:57working. I've chosen a strategy on

5:59Hyperliquid. This is an example strategy

6:01for this video that I came up with with

6:03Claude, but what you could do is you

6:04could backtest your own strategies and

6:06you could implement them into the system

6:08that I'm going to give you. So, I'll

6:08show you how to do that in a second.

6:10But, essentially this is a real market.

6:11This is the Hyperliquid price right now.

6:13You could do this on Bitcoin. You could

6:14even do this on stocks. I'll show you

6:16how to connect this to a live account

6:17very shortly. And essentially every

6:19single dot is Jev making a decision. So,

6:21it's either a buy decision or a sell

6:23decision. And Jev is looking at three

6:25criteria before making every call. Who's

6:27buying? Who's selling? Which way price

6:29is moving? And you can train based on

6:31what strategy you give it the

6:32prerequisites or the questions that you

6:34want Jev to ask in order to assign a

6:37probability and then make a buy

6:38decision. I think this has applications

6:40beyond high-frequency trading like this.

6:42I think for investing it could also be

6:44valuable because you could essentially

6:46trigger Jev to make an adjustment to

6:48your portfolio if there is a variable

6:50that changes. For example, smart money

6:52starts exiting. You could have a

6:54specific investor type or wallet type if

6:56it's an on-chain asset trigger a

6:58probability that changes your material

7:00position if Jev picks up on it. Or you

7:02could have momentum-based criteria. If

7:04momentum is moving to aggressively in a

7:06certain direction or we have a

7:07volatility regime shift, it could make a

7:10tweak to your notional position sizing.

7:12So, the possibilities are really

7:13endless. Hopefully this video opens your

7:16eyes to what you can do at least on the

7:17high-frequency trading front, but I

7:19don't want this to limit the potential

7:21of what you could do with a tool like

7:22this. I mean, we're making decisions as

7:24you can see here in like 300, 400, 500

7:27milliseconds. This is extremely quick.

7:29It's way quicker than using Claude as

7:31the decision-maker using its inference

7:32cuz it literally can take 5 seconds.

7:34Like, the difference in high-frequency

7:36trading, in news trading of a few

7:39seconds is astounding versus like 400

7:41milliseconds, 300 milliseconds, which is

7:43why this is so important because you're

7:44compressing the speed down. I think the

7:46application for news trading is massive

7:48as well. As you can see in front of you,

7:49I have a news dashboard, which is

7:50literally getting live news right now.

7:52It's assigning it a probability, and

7:54then Jev is working out on a

7:55probabilistic basis, is it bearish or

7:57bullish? And then you could have this

7:58automate a particular positioning either

8:01on a scalp trading basis like the

8:03example in front of you or on a even

8:05high time frame basis in terms of high

8:07time frame portfolio management. I'll

8:09circle back to this in a second. So,

8:10once you have the technical

8:11infrastructure set up, what you need is

8:13a strategy. As I said before, I'm

8:15running a paper trading strategy, which

8:16was based on some rough logic, but if

8:18you want to find a strategy that is

8:20going to work well with this, you need

8:22to do backtesting. So, step two, after

8:24you have Jev set up using the one prompt

8:26guide, is to go and find a strategy.

8:28Now, there are many ways to do this. I

8:29have other videos on this. There are

8:31many free available tools that you can

8:33use to find strategies, but what I'll

8:35typically do is I'll use a backtesting

8:37engine. So, this is actually one that

8:38I've built myself, and maybe we'll

8:40release it at some point in the future,

8:41but right now I'm just using it for

8:42myself, where I can actually design

8:44backtest. So, I can design a backtest on

8:46Bitcoin, for example. It will recommend

8:48me strategies. I can tell it the type of

8:49strategies I'm looking for, and then it

8:51could run multi-hour backtests to go and

8:53find a bunch of strategies, and then I

8:55can put them into TradingView, and I can

8:57try and filter the ones that look

8:58legitimate, and then I can start running

9:00forward tests on them. Once the forward

9:02tests come back profitable in the

9:04out-of-sample period, then what I can do

9:06is I can set it up with Jev and start

9:08paper trading. Once I've done paper

9:09trading for a certain period of time,

9:11and I'm confident the strategy is going

9:12to work, then I can make it a live

9:15trading strategy with real money, and

9:16I'll show you how to do that in a

9:17minute. The reality of strategy

9:19selection is that there's no magic pill

9:21for this. It is 99% of strategies lose

9:24money. It is possible to find profitable

9:26strategies, but what you need to do is

9:28run hundreds or thousands of backtests,

9:30which can be automated with a

9:32backtesting engine like this. You need

9:33to find a strategy that looks good. You

9:36can ask Claude and use these providers.

9:38That's what these models are really good

9:39for, the inference side of things, to

9:41determine why a strategy's working, why

9:44it's good, why it's profitable, and

9:45what's wrong with it, refine it, put it

9:47into TradingView. This is what I do lots

9:48of videos on, so make sure to subscribe.

9:50This video is mostly focused on Jev. The

9:52strategy hunting part I have many other

9:54videos on, and I'll continue to teach

9:55this in my school community, so make

9:57sure you join my school community, as

9:58well. And then once you find outlier

10:00strategies, you can then start running

10:02them in a live environment. Now, the

10:04problem with high-frequency trading,

10:05specifically for Jev, is that the fees

10:07can chop you up, because remember you

10:09are paying a fee, a maker-taker fee,

10:11every time you execute a trade. The way

10:13to think about Jev is it's like a coach

10:15that's on the sidelines. All it's doing

10:17is defining a probability and screaming

10:19out, "Buy! Sell!" It's basically just

10:22determining how short it is. But the

10:23code that you write, the strategy that

10:25you come up with, with the LLM, that is

10:27the rulebook that the game is being

10:29played on. Now, Jev doesn't cost money,

10:31so, you know, if it shouts at the

10:32players, that's not actually costing you

10:34money. You can get it to give as many

10:36probabilities as you want. What's

10:37costing real money is the strategy side

10:39of things. That's why you want to do the

10:41backtesting beforehand, and then you can

10:43set up a sustainable strategy that Jev

10:45can just implement. And I know some of

10:46you might be watching this video, like,

10:47"Okay, Miles, just give us a profitable

10:49strategy." The reality is that you need

10:51to test things out yourself, because

10:52there's no universal strategy which is

10:54going to make everyone money. Now, at

10:56the moment I'm running multiple forward

10:57tests. Some are losing money, and I'm

10:59having some that are exhibiting great

11:00results. I'll make videos on them when I

11:02feel like the time is right. What I

11:04would tell you today is set this system

11:06up, follow the steps that I've showed

11:08you, and I'm about to show you how to

11:09make it live. Go use a backtesting

11:11engine. If you don't have access to a

11:13custom engine, you can go into Claude

11:15code, you can ask it to create a

11:16backtesting engine harness, and you can

11:18get the latest models, for example, Opus

11:205.5, which is a very smart model to

11:22start backtesting strategies. You can

11:24also find strategies. I did a whole

11:25video on this, I'll link it below, on

11:27how to find strategies on Quantpedia and

11:29NFX, and then you You look at the

11:30out-of-sample results, and then you can

11:32paper-test the forward results, and

11:33start paper-testing the stuff with

11:35today's system. So, my number one

11:36recommendation would be to not actually

11:38concern yourself with the strategy too

11:39much. I think a lot of people, they want

11:41to find the ultimate strategy, but

11:43really what you need is the

11:44infrastructure first. So, learn how to

11:46backtest strategies, grab a strategy,

11:48and then implement it on paper trading,

11:50so not real money, so you can actually

11:52learn how to do it, and then if it has

11:53results on paper trading, then connect

11:55it to real money. Something that I built

11:57that I haven't seen many people using

11:58Jev for, but it's a great application of

12:01it, is the ability for Jev to decipher

12:03real-time news headlines. So, if you

12:05feed it a Bloomberg, a Trivela, a

12:07Reuters, you know, you can get free

12:09sources like Yahoo Finance. The problem

12:10is that it won't be as fast, so you're

12:12already eroding edge there. So, Jev is

12:14like helping you speed up edge on

12:16automation. So, you also want to have

12:17like a bit of edge on the feed as well.

12:19So, you have to plug in a fast news

12:20source. The more you pay, the faster the

12:22news source is going to be. You have Jev

12:23reading that. It does it in a fraction

12:25of a second. It's asking, all right,

12:26what coin, stock, or asset is being

12:29referenced? Is it going up or down? How

12:31big should we size? Is it bullish or

12:32bearish? And then you can use that to

12:34automate execution with preset

12:36parameters. This is a very fun test that

12:38I'm running right now. I've been running

12:40it for the past couple of days, but it's

12:42going to take, I think, a couple of

12:44months for me to get the real data as to

12:46whether you can develop a strong edge

12:48news trading with Jev. My hunch is that

12:49you can if it's optimized correctly, but

12:51just like I said before, these

12:52strategies need refinement. These aren't

12:54just things that you can just like set

12:56up overnight. Oh, it's a magic money

12:57printer. Nothing's a money printer.

12:58Discretionary trading also takes a long,

13:01long, long time, arguably longer to

13:02master. So, I think it's good to be

13:04realistic with you guys about what AI

13:05trading actually is. It's not

13:07get-rich-quick, but if you can find the

13:09right strategy, you can make sustainable

13:11edge over time. And it's a compounding

13:13game. I have some assets in my AI

13:15trading bots. I have some assets that I

13:16trade with myself, so discretionary

13:18trading. I have assets that I buy and

13:20hold longer term. Diversification is

13:21always the best approach, but the reason

13:23I'm going so deep on trading bots is

13:25because I believe by being at the

13:26forefront of this, the edge that is

13:29going to be created over the next couple

13:30of years is monumental considering how

13:33good this technology is getting and so

13:34many people just don't understand how to

13:36apply this to investing and trading yet.

13:38That in itself is an edge, whereas the

13:40markets overall, I find are already so

13:42arbitrage, it's much harder to get edge.

13:45So, that's the entire reason I'm diving

13:46really deep into AI, but you can see in

13:48front of you, this is the newsroom that

13:49I created and this is the system that I

13:51just showed you in action. So, you have

13:52the headline coming in, you have Jev

13:54reading the headline, assigning whether

13:56it thinks it's neutral, bullish, or

13:57bearish, and then occasionally a trade

13:59signal will fire. And it's doing it

14:01extremely quickly because it doesn't

14:03have that chat interface inference, it's

14:05just assigning a probability and you're

14:06good to go. So, that's essentially how

14:08Jev works. It's assigning a match based

14:10on a preset set of variables that you

14:13will define with an LLM prior. And this

14:15is why Jev is actually really good in

14:16business as well because if you're

14:18scanning a lot of data, so if you're

14:19doing SEO or you're doing business

14:21development, it can take a lot of data

14:22in and it can assign a match to a

14:24prospective client, to a piece of

14:26information on the internet, and it can

14:28label and categorize that data really

14:30quickly and it costs you a lot less to

14:32do so. So, anything that requires mass

14:34data consumption and mass

14:35decision-making is where you want to be

14:37using Jev. So, now for the fun part. How

14:40do we actually connect and exchange so

14:42you can start trading? Well, you've got

14:44two parts, I'll go through both. The

14:45first one is on Alpaca. So, Alpaca has

14:48paper trading, it also has live trading.

14:50You have crypto, you have stocks. It is

14:52slightly higher fee, but because you can

14:54access multiple asset types, it is quite

14:56convenient and I'll be doing lots of AI

14:57experiments on Alpaca. You simply want

14:59to take your key, your API key, and this

15:02will all be part of the setup prompt,

15:03put it into Claude code when you're

15:05setting it up, and put it into the .env

15:07file that Claude code will prompt you to

15:09set up to connect your API key, and then

15:11Claude, which is the brain in this

15:13instance, will feed the data to Jev and

15:15it will ping a signal to your paper

15:17trading account in this case, and you

15:19can switch it to live trading once you

15:20want to use live trading. It's a simple

15:22way to do it on Alpaca, it's all part of

15:23the setup prompt. Path number two is

15:25more complex, but more sustainable if

15:28you're actually trying to run this at

15:29scale. So, path one is good if you want

15:30to test it out. Path two is operating an

15:32always-on server, so you could go onto

15:34Hosting or for example, get a VPS. I've

15:37got a discount for Hosting or below if

15:38you ever want to start a VPS, by the

15:39way, it's better than just going through

15:41the website, helps out the channel, but

15:42also gives you a discount. That

15:43essentially enables you to get an

15:44always-on server because the problem

15:46with Claude code is that the second your

15:47computer goes to sleep, it's no longer

15:50doing inference. Now, Jeff technically

15:52still is because Jeff, in my case, is

15:54deployed through a reversal, so that is

15:56in the cloud, but if you need inference

15:58from Claude on any decision-making, then

16:00you're going to want an environment

16:02where it's always-on. So, that's why I

16:03recommend having a VPS. Or you can use

16:05a, you know, if you have a Mac Mini or

16:06you have a local PC that can always be

16:09on, then that's going to be great

16:10because then you can just have a session

16:12that never closes. That way you can

16:14connect it to Claude as the brain, and

16:16then the brain can fire off signals to

16:18an exchange. In this case, an exchange

16:20could be Bybit. I believe I also have a

16:22special deposit bonus down below if you

16:24want to sign up for an exchange account.

16:25Bybit has pretty much every crypto, much

16:27better depth and lower fees than Alpaca,

16:29alongside commodities. I think it has a

16:31lot of perp stocks now, so that's

16:33actually my preferred option to Alpaca

16:34in in many cases. I think they also have

16:36paper trading, so you can connect it

16:37directly through their MCP. You just

16:39need to connect your API to Claude. The

16:41other thing you can do is you could use

16:43in between Claude and Bybit, you could

16:45have something like Trigger Trade, which

16:47is just triggering it. That's a website

16:49that enables you to just trigger based

16:50on a Pine Script in the cloud, which

16:52reduces the need for a VPS because you

16:54just give it the Pine Script, it just

16:55executes for you on Bybit. But, if

16:57you're trying to use Jeff, then that's

16:58not as effective because the idea is to

17:00have an integrated autonomous system

17:02that you have sovereignty over. But, the

17:03reality is if a strategy is really just

17:06plug-and-play and it's already been

17:07verified using something like Trigger

17:09Trade is great. If you want more

17:10sovereignty over it, and if you have a

17:12strategy that's more adaptive, or you're

17:14just running a test to see, you know, if

17:15Claude can self-adapt, then you're going

17:17to need like an always-on device. So,

17:19these are the two paths. They're both

17:21outlined in the setup prompt, so it'll

17:22help you set them up. The other

17:23important thing to set up are risk

17:25rules, so max position, daily loss, kill

17:27switch, and potentially manual approvals

17:30for certain types of trades. You can

17:31also set up an alert system. This is

17:32what I do. So, on my iPhone, I get

17:34notifications when a trade enters or

17:37exits. Obviously not high frequency, I'd

17:39be getting spammed, but especially on

17:40the lower frequency, higher timeframe

17:42stuff, I'll get alerts, so I can review

17:44things. On these strategies where I'm

17:45using more money, so bigger orders, I

17:47will actually still do manual approval.

17:49So, I'll get a message on Telegram,

17:51it'll say, "Are you okay to go through

17:52with this trade?" I check it quickly, I

17:53say yes. This is for like big high

17:55timeframe swing when there's big stakes

17:57involved, cuz I know people, you know,

17:58are still at the stage, especially if

18:00you're testing out a strategy, if you've

18:01just gone from the paper forward test to

18:03a live money test, sometimes, especially

18:05if it's big high timeframe stakes, you

18:07don't trust the LLM to do it 100%. I

18:09think we're going to get to a place

18:10where we do, but that's not the reality.

18:12Hallucinations still exist, so you can

18:14have that fail-safe still. But, the tech

18:16is getting a lot better, and it's so

18:18much better than it was just a year ago.

18:19So, that's it. That's how you can use

18:21Jev in conjunction with an LLM like

18:24Claude to do some pretty amazing things

18:26when it comes to low latency trading. If

18:28you want the setup prompt, it's in the

18:30school community. Just go into the

18:31classroom, go to the free assets

18:32library, claim it, drop it into Claude.

18:34Let me know in the comments if it worked

18:36for you, if there's anything you want to

18:37change. We'll read the comments, and

18:38we'll make adjustments to it to ensure

18:40that it continues to improve over the

18:42coming weeks. Shout out to Rowan, shout

18:43out to Lewis for their inspiration for

18:45this video, and I will see you in the

18:47next one. Have a lovely rest of your

18:49day. Peace out.