Algo trading research with Google’s Antigravity and Gemini 3 — Transcript
Full transcript
- 0:00Not only Google just released a Gemini 3
- 0:02model, they also released an agentic
- 0:04IDE. And even though there's no way I'm
- 0:06going to get used to that name, I'm
- 0:08actually super excited about this thing
- 0:09because it's not just for coding, it can
- 0:11also control the browser for you. And
- 0:14yes, at this point Cursor does that too,
- 0:16but this one actually does it better.
- 0:17The other thing is that if you want to
- 0:19use so many tokens in your Cursor IDE,
- 0:21you're going to have to pay up to $200
- 0:23per month, but this one is completely
- 0:26free, at least for the [music] time
- 0:27being. However, in this video, I want to
- 0:29show you how it up specifically for algo
- 0:31trading so that it can do research for
- 0:33us, to write a strategies, to modify
- 0:35them, to even run them, and look at the
- 0:37results, and maybe try different
- 0:39combinations of indicators or things
- 0:41like that, and find us the best
- 0:43combination. So, I couldn't be more
- 0:44excited about it, so let's get right
- 0:46into it. So, once you install
- 0:47Anti-Gravity, you're going to get this
- 0:49interface. Now, it took me a while to
- 0:51import my settings from Cursor. For some
- 0:53reason, my shortcuts weren't working, at
- 0:56least some of them, and I had to
- 0:57manually change them. So, that was a
- 0:58little bit annoying, but it is
- 1:00definitely worth it. Now, once you're
- 1:02done, you will get a typical agentic
- 1:04IDE, right? But, here's the part that
- 1:06you need to pay attention. To ensure we
- 1:08can use this thing for algo trading,
- 1:11we're going to have to feed the model
- 1:12some instructions. If you remember this
- 1:15video, which I made like 6 months ago,
- 1:17in it I explained how to feed the custom
- 1:19instructions into any kind of LLM model
- 1:22so it can give you output specifically
- 1:24for algo trading in Python with the
- 1:25Jesse framework, of course. But now, my
- 1:27goal is to be able to tell the IDE to,
- 1:30"Hey, write this a strategy for me." And
- 1:32it should do it exactly there, so I
- 1:34won't have to copy and paste code
- 1:35anymore, and hopefully it should even be
- 1:37able to run the browser for me, and
- 1:40maybe execute the backtest for me, or
- 1:42look into the results. Now, in that
- 1:44video, I open-sourced the instructions.
- 1:46Now, here's the repository, and I'm
- 1:47going to post the link of it in the
- 1:49description of the video. Now, this is
- 1:51the file that you need, instructions.md.
- 1:53So, let's open it and copy the whole
- 1:55thing. Then, inside your IDE, create a
- 1:57new file called agents.md, and in it
- 2:00simply paste everything. Now, by the
- 2:01way, this is a new Jesse project, and in
- 2:03case you forgot how to generate a new
- 2:04Jesse project, just go to the
- 2:06documentation, this page, and this is
- 2:08the command. So, just copy this, and it
- 2:10will generate a new Jesse project for
- 2:11you. Then, you also need to copy the
- 2:13.env file, which I did, and that's it.
- 2:15Now, to run this project, all I need to
- 2:17do is to say Jesse run. Now, of course,
- 2:19you need to ensure you are running the
- 2:21correct conda environment so that it
- 2:23should have the correct Python version
- 2:24for you. So, let's do this.
- 2:26There we go. It's running it at this
- 2:28URL. For you, this will probably be
- 2:309,000. I have changed the default port.
- 2:33So, let's just copy this and open it
- 2:35inside our Chrome browser. Now, the
- 2:36default password is test. There we go.
- 2:38So, let's go to the strategies section.
- 2:40Now, here we have the option to import
- 2:42one of the existing strategies from
- 2:43Jesse's community, or we can create a
- 2:45new one. So, let's create a new one, and
- 2:47I'm going to name it test one. All
- 2:49right, so this is the boilerplate that
- 2:51we get when generating a new strategy.
- 2:53But, instead of coding it here, I'm
- 2:54going to open Anti-Gravity, and let's
- 2:57open it here. All right, so I can close
- 2:59this now. Now, here we have the option
- 3:02of planning and fast. For these tests,
- 3:04I'm using the fast option, and for the
- 3:06model, this is the most intelligent
- 3:08model right now, Gemini 3 Pro High. But,
- 3:11I'm setting it on low because I found it
- 3:13to be good enough for these type of
- 3:14tasks, and it is significantly faster.
- 3:16But, if there is a point that it is not
- 3:18doing what you need, then set it to
- 3:20high. It also gives us unlimited access
- 3:22to Claude Sonnet 4.5, which is
- 3:24absolutely amazing, and I did test it,
- 3:26and it works great. But, let's go with
- 3:27Gemini 3 Pro for this test. All right,
- 3:29so let's keep it simple. Write a golden
- 3:31cross strategy. But, to make sure that
- 3:33it knows we want to edit this file, I'm
- 3:35going to attach this file. Let's hit
- 3:37enter.
- 3:43There we go. So, it defined a short
- 3:44moving average with a period of 50,
- 3:46another one with 200, and here it is
- 3:48ensuring that a crossover is happening.
- 3:50Now, this isn't how I usually write a
- 3:53golden cross. I just keep it simple
- 3:54without the sequential true so that we
- 3:56don't have to select the last item. But,
- 3:58this should work just fine as well, and
- 4:00it also defined the position sizing, and
- 4:02everything looks fine. It's also saying
- 4:05that once those positions are no longer
- 4:07valid, you should liquidate the current
- 4:08position. So, this is perfect. Good.
- 4:10Let's accept it, and let's open the
- 4:12dashboard again, go to the backtest
- 4:14page. Let's change this test one
- 4:17strategy, choose the 15-minute time
- 4:19frame. BTCUSDT is fine, and the Binance
- 4:22Perpetual Futures exchange is also fine.
- 4:25This period is is also fine for this
- 4:26test. Let's enable the fast mode, and
- 4:29let's give it a go.
- 4:31And there we go. So, it runs perfectly
- 4:33fine. It made 22% max drawdown of minus
- 4:3711 in the period of 2 months, and a
- 4:39sharp ratio of three. So, of course, we
- 4:42are cherry-picking the period here, but
- 4:43it seems to be working really well. So,
- 4:45next, I want to see if the model is
- 4:47smart enough to add some kind of filter
- 4:48based on the condition that I am
- 4:50describing to it. The strategy is good,
- 4:52but it is taking too many trades. Can
- 4:54you define some kind of filter for it to
- 4:56not take trades in ranging markets?
- 5:00Let's see how it does. Now, one issue
- 5:03that I'm facing with this editor is that
- 5:05I find it to be a little bit slow, like
- 5:07with the AI tasks, and maybe this is
- 5:09because they launched it just yesterday,
- 5:12and they are definitely under heavy
- 5:13traffic, especially because it is
- 5:14completely free right now. But, overall,
- 5:16I found Cursor to be faster. Now, by the
- 5:18way, you can definitely use Cursor to do
- 5:20exactly the things that I'm doing in
- 5:21this video, but it won't be free, and
- 5:24also, for the last part where I I'm
- 5:25going to ask the IDE to control the
- 5:28browser, I don't think Cursor can do it
- 5:30as good as this one. All right, so it
- 5:31defined an ADX indicator. This is the
- 5:34correct syntax. Yes, it is, and in the
- 5:38should long method, it is adding this
- 5:40condition now. So, this seems perfectly
- 5:43correct. So, let's accept it, go back to
- 5:45the dashboard, and rerun it one more
- 5:47time.
- 5:48And yeah, it works. All right, so it
- 5:50works, but it made the results worse,
- 5:52right? So, here's the next part that I
- 5:54wanted to do. So, let's go back and
- 5:57actually return the code to what it was,
- 5:59but this time reject it, and let's open
- 6:02a new conversation entirely. Just give
- 6:04it this one one more time. Now, I want
- 6:06to see if it's able to add a new
- 6:08condition, test it with the browser all
- 6:10by itself, and if it's not better, which
- 6:13it wasn't in this case, it needs to be
- 6:15smart enough to roll back the previous
- 6:17code and tell me about it. So, in other
- 6:19words, I want to see if this thing is
- 6:20actually able to do research for us. So,
- 6:22maybe I want to keep running it for 20
- 6:25or 30 minutes, right? So, keep trying
- 6:27different combinations of indicators and
- 6:28things like that. So, we're going to
- 6:30give it a simple test. If it passes this
- 6:32one, I'm pretty sure it can also do
- 6:35bigger things. So, this is the last
- 6:36thing I needed to work. The strategy is
- 6:39good, but it is taking too many trades.
- 6:40I need you to add an ADX filter, and
- 6:43then open the browser and run the
- 6:45backtest. If it improved the results,
- 6:47keep it. If it didn't, remove it.
- 6:50Now, we also need to tell it what's the
- 6:52URL of the dashboard, right? So, let's
- 6:55say the URL is Let's open Jesse. Give it
- 6:58this exact URL, but also we can restart
- 7:02a new session. Let's go back and paste
- 7:04the URL. Now, before I hit enter, I got
- 7:07to say that the first time that you do
- 7:08this, it's going to ask you to set up
- 7:10some kind of extension, and you should
- 7:12allow it to do that. Now, I already did,
- 7:14that's why it says remove from Chrome
- 7:15now. Now, the other thing is that
- 7:17sometimes when it opens the dashboard,
- 7:18it needs the password, right? So, let's
- 7:20also give it that. The password,
- 7:23if needed,
- 7:24is
- 7:26test.
- 7:29All right, so it's starting working, and
- 7:31it is starting to navigate inside the
- 7:34browser. I'm going to say always allow
- 7:37so that it can always open the browser
- 7:39automatically, and let's open the
- 7:41browser. There we go. So, you see, it is
- 7:44seeing the password page now.
- 7:47It entered the password.
- 7:49Let's give it some help here.
- 7:52Now, I'm going to fast-forward the video
- 7:54a little bit because it can be slow
- 7:56sometimes.
- 7:57All right, so it reran the strategy.
- 8:02Let's go back to the IDE, see what it's
- 8:03saying.
- 8:05So, the backtest is running.
- 8:09Change the code.
- 8:13All right, so it just implemented the
- 8:14ADX. Let's see what it does now. Now, by
- 8:17the way, whenever there's this blue
- 8:19color here, we can know that the agent
- 8:21is controlling the browser. So, that's
- 8:23pretty helpful.
- 8:24It clicked on the rerun button. It's
- 8:27seeing the new results now. So, let's go
- 8:29back to the IDE.
- 8:33The ADX filter reduced the number of
- 8:35trades from 38 to 14, but it also
- 8:37significantly worsened the net profit
- 8:39from 22% to minus 4%. Based on the
- 8:42instruction, "If it improved the result,
- 8:45keep it. If not, remove it. I should
- 8:47remove the ADX filter. I will now revert
- 8:49changes made to the file." It says,
- 8:51"I've completed task. Here's the summary
- 8:53of the experiments. Baseline backtest,
- 8:55here's the result. With ADX filter,
- 8:57here's here's result. Although the ADX
- 8:59filter successfully reduced the number
- 9:01of trades, it significantly hurt the
- 9:02profitability, turning it negative. As
- 9:05per your instructions, to only keep it
- 9:07to improve results, I have reverted the
- 9:09changes to whatever it was." So, this is
- 9:12fantastic. There we have our answer.
- 9:14Yes, this thing can do research. Now, I
- 9:16know it took a while to finish the task
- 9:18for now, but imagine that you write this
- 9:21thing at night, you let it run, and then
- 9:24you go to bed. You wake up tomorrow, and
- 9:26it has tested maybe 100 different
- 9:29conditions for you. And right now, all
- 9:31of this is completely for free, so it
- 9:34couldn't get any better. Now, there's a
- 9:36lot more we can do with it. If you like
- 9:37this video, please let me know in the
- 9:38comments, and let me know if there's any
- 9:40kind of experiments you want me to do
- 9:41with it, and maybe record another video
- 9:43about it. Now, we are going to have a
- 9:45giveaway. A random person who likes this
- 9:46video, posts a comment, and subscribes
- 9:48to the channel is going to win 1 million
- 9:50bong token. All right, let's pick the
- 9:51winner for the previous video.
- 9:56And the winner is, Jesse. Thanks. Could
- 9:58you make this strategy for hard FX
- 10:00pairs? Now, my name isn't Jesse, it's
- 10:01Solo, by the way. Thank you so much for
- 10:03your comment. Please reach out to me so
- 10:04I can send you your bank tokens. Thank
- 10:06you so much for watching. I'll see you
- 10:08in the next one.
About this transcript
This page contains the full transcript of Algo trading research with Google’s Antigravity and Gemini 3 by Algo-trading with Saleh, generated from the public captions YouTube serves with the video. The transcript has 2,083 words across 292 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.