I Turned GPT-6 Astra Into a 24/7 Stock Trader (tutorial)

0:00 So, I recently did a trading challenge where I gave Claude $10,000 of my real money to trade stocks. And at the end of the month, I was actually beating the S&P by a little over 8%, which I was

0:08 pretty impressed by. And now that GBT6 Astra is out and it's even more powerful, I'm going to do the exact same thing. So, before I kick off that $10,000 challenge, I wanted to actually

0:16 just show you guys how I'm getting this all set up so that you can replicate this if you want to turn Astra into a 24/7 trader for you. So, let's not waste any time and just get straight into this

0:23 one. All right, so before we get started, I just want to say this is not financial advice. I'm not recommending you give an agent $10,000 to trade. I'm recommending that maybe you start with

0:31 paper trading, you get a strategy that you like, and then you automate some of that. Or if you're already consistently trading every day and every week, then maybe just see if you can use AI to help

0:39 you get more updates or to help you keep an eye on things so that you're not having to like pocket watch different stocks or something like that. But anyways, let's just jump in here and let

0:46 me show you how simple this is to set up. So right here where I'm at is I'm about to set up all of the automatic routines so that this thing can run without me. So, I wanted to show you

0:54 guys, I asked right now in as simple terms as possible, explain our strategy and what we're going to do. So, here's what Astra came back with. We built the strategy around three constraints.

1:03 $10,000, seven trading days, and we're doing six wakeups or six checks per day. So, what I started off with was I literally was just having a conversation with Astra. I was saying, "Hey, I have

1:12 this challenge. I want to take, you know, a new strategy, so it's a little bit more risky. I have seven days to try to make money." And I asked it to just do a bunch of research. So, it fanned

1:20 out like 10 sub agents, and it researched a ton of stuff. As you can see, we've got this big dock where we talk about the strategy and it consolidated a bunch of research

1:26 information. Now, this is where I see there being sort of two paths. You have the path where you don't really have a trading strategy and you're wanting to sort of outsource that research and

1:34 thinking to AI agents or you've got the path where you are consistently trading every day and you want to just help yourself be able to move faster or not, you know, forget things. So, in this

1:43 case, what I did is I wanted to outsource that thinking and I wanted to have Astra come up with a strategy. I'm basically just helping guide it. And what it came up with is this schedule.

1:51 So 7:45 a.m. Central, we're going to read the news, check the account, and choose stocks to watch. So, let me just pull up a little calendar here to show you guys what that actually looks like.

1:59 Here's the first time that Ash is going to wake up on trading days about 7:45, and it's going to read the news, check the account, and choose what stocks to watch. Now, this is where I would say if

2:08 you are someone who's consistently trading, think about your day like this, you know, market open, market close, what do you do? Maybe before the market opens, do you look at some stuff? Maybe

2:17 an hour in, do you look at some stuff? you know, maybe before close, do you look at some stuff? And just sort of put yourself on this like 1-day calendar because then you can just basically make

2:25 agents to do those things at those times to help you out a little bit. But anyways, let me just keep going here with the proposed plan from Codeex. At 9:30 a.m., it's going to look for the

2:34 first qualifying trade. So now we have agent wake up number two. At 11:00 a.m., it's going to review positions and consider the last new trade. At 1 p.m., it's going to manage existing positions.

2:43 At 2:15, we're going to close out the remaining positions. And so it's kind of taking more of a day trading strategy rather than a long-term trading strategy. And then at 245, right before

2:52 a close, it's going to confirm we're out and record the day results. So that's how simple this is going to be. Every single trading day, it's going to do these actions at these times. And that's

3:01 how we're going to basically make this 24/7 trader for our 7-day challenge. Now, if you'd feel more comfortable having another wake up after the market closes, you can do that. Or if you want

3:09 another one before the market opens, you can do that as well. This is just what Astra came up with for this specific challenge. Now, the important piece that I wanted to call out here is about

3:16 continuity. Meaning, every single time that one of these agents wakes up, it's stateless. And so, what we have to make sure it does is that it reads through the project, it looks at our portfolio,

3:25 and it understands what it needs to do. It understands the context of what just happened in order to understand better what I need to do now. And so, the way that I'm kind of tackling this here is

3:33 having every single wakeup end with a progress log. So, that way we have basically this system where, you know, one agent wakes up and starts working. It will basically say, you know, here's

3:42 what I did today and here's what comes next. So then when this next agent wakes up and basically takes over, it will read that output. It will do its work and then go, okay, cool. Here's what I

3:51 just did. And then we just basically have this loop where every single time the next agent wakes up, it's able to take over all the work. It's able to orient itself with what's going on and

3:59 just keep actually feeling like it's one trading agent rather than having a bunch of different isolated workers. So you can see here it says continuity comes from shared records, not the AI

4:08 remembering the conversation. So every wakeup will read the strategy and the previous handoff, check Alpaca, which is where we're trading. I'll show you guys that in a sec to see what actually

4:16 happened, perform its assigned job, update the current account state and progress log, and then leave clear instructions for the next wake up. And if you don't have this continuity, it's

4:25 just going to feel like you're just kind of like throwing darts at the wall. Critical actions get recorded as they happen, so an interrupted run can be recovered without accidentally repeating

4:33 a trade. So all of this stuff is ready and now we just have to actually set up the connection to Alpaca and set up the actual routines. And setting up the routines is important. So I will talk

4:41 about that when we get there. But I did real quick want to show you guys some stuff in this doc. So here are the persistent files. The ones I wanted to call out are the progress log. You can

4:49 see here's everything that's going to happen. We have an actual journal as well which can be appended and we have evidence. Now I also brainstorm with Astra on some failures. So failure

4:58 behavior would be no previous progress record on the first run. Next model has no chat history. prior run crashed after sending an order, two runs overlap, missing or corrupt, or the machine or

5:07 app go offline. And by the way, guys, to make this easier, I did put all of this into a free resource guide so that you can basically give this to an agent if you want to help it sort of set up the

5:15 same sort of system that we have here. And you can get this for completely free by joining my free school community. The link for that is down in the description. You'll come in here, you'll

5:21 go to classroom and click on all YouTube resources, and everything will be in there. And you'll find the doc associated with this video and all of my other skills and repos and things like

5:29 that. So, if you want that doc, go ahead and grab it. But let's get back to the video. So, next up here is we need to go to Alpaca to get the actual connection set up. So, here is my Alpaca account.

5:38 You can see that if you make an account here, you can also have a paper trading account. So, you could get on here and connect your agent to this and just have it trade for you for, you know, a couple

5:47 months or a couple weeks or however long it takes for you to feel comfortable with the strategy. And then you can move it over to real money. But this is just paper trading. So, it's going to follow

5:54 the market. It's real stocks, but it's fake money. So, you can paper trade first and then you can open up an actual brokerage account here. And this is $10,000 of real cash in this account.

6:02 Now, so if you want to actually follow along with this sort of video, then you're going to want to go get an alpaca account set up. So you can grab the API key and put it into your codeex. You can

6:10 see right here there's API keys. It gives you this endpoint. And then we have the actual key right here. So what I'm going to do is I'm going to copy this key. And obviously this is like a

6:18 password to your trading account. So don't give this key to anybody. Also, don't put this key into the chat window of CEX. I'm going to show you where you put this. What you're going to do is

6:26 you're going to ask it to create you aenv file or if you're already in your AOS or if you're already in a project that has one, you're going to put it in there. Now, what I did specifically is I

6:34 put this trading challenge inside of my AIOS. So, if I open up these files, you can see this is what my project for this trading challenge looks like. We have our docs, we have our agents.mmd. We

6:44 even have a cloud. MD if we want to move it over. We have scripts, routines, we have ourv, but this actually lives within my HERK 2. So, it's in my Herk 2. It's in a folder called other worlds,

6:53 which is where I have a bunch of other worlds. And then we have our Astra trading bot challenge. So, I'm going to be running the routines in this project just so I can keep everything sort of

7:01 like isolated, and I don't want too much context polluting the window. I don't want confusion or bloat to get into my trading bot because all my trading bot has to think about or look at is trading

7:10 information. It doesn't need to know about my business or my projects. But, I do want my overall herk 2 to be able to have better visibility into the trading challenge just for context. So, that's

7:19 how I'm isolating this project. So anyways, what I'm going to do is I need to open up my files over here and I need to put my alpaca API key that you guys just saw me copy. I need to put that

7:29 into theenv. I'm going to go to the Astra trading challenge and I'm going to open up theenv and I'm going to paste it right in here. Now, when you do this with real money, you're going to have to

7:36 click on regenerate because that will give you an API key and an alpaca secret. And you do need both of those things. So you'll copy the key and then you'll put that in the right spot in

7:44 here. You can see right here, a alpaca API key. And now you can see that I've pasted in my API key as well as my secret. And I save that. And now we want to check if Codex can actually look into

7:54 that account. All right. So I just gave you those keys for Alpaca. Can you just make sure you can actually see in the account and that you understand the end points and everything like that so that

8:01 you can take action in there. There we go. So Alpaca reports your live account with $10,000 cash and equity. There's no positions, no open orders. And now what else I did inside of my account here is

8:10 if I go to the um plans and features, if we go to market data right here, we can see that I am on a plan for 99 bucks a month to get like real-time market coverage and I get extra API calls per

8:21 minute and I get just more data in here. Now, I'm not saying you have to do this to start. When I actually did my 10K trading challenge with Claude, I didn't pay for this. So, it was basically just

8:29 doing research on the web and using like secondary research in order to make the trades. But because this one is 7 days, I probably want more real-time market coverage. So, paying for a subscription

8:39 like this for a month just to be able to get better data for this challenge is going to be important for me. But like I said, that's not necessarily something that you have to do. And what else is

8:46 cool is if you go to the plugins over here and you search for something like Alpaca or you search for stocks, there's a bunch of other things you can connect. So right here, you can connect Alpaca,

8:54 but this typically gives you market data rather than letting you trade through it, which is why we do the API key. But there are also tons of other ones. There's alpha stocks, there's massive,

9:01 there's stock and ETF research panel, there's trading cursor, there's tons of other, you know, subscriptions you can get on or platforms that have other types of trading data or, you know,

9:09 technical analysis that you can connect to codecs for better research. So anyways, now you can see that Astra came back and said, "All of the keys work and I do understand how I can trade and how

9:18 I can use your buying power and I can do everything." And the next regular open is tomorrow, September 8th, because today is Labor Day, so markets are closed. So now that all of this stuff

9:27 has been confirmed and we need to start getting everything set up for tomorrow, we have to go ahead and start building that routine out or that scheduled task. Now here's the one distinction about

9:35 scheduled tasks. There are two different types. There is a local one or there is a cloud one. And so if I come here and I click this device, this means it runs locally. And actually for this challenge

9:46 specifically, we are going to keep it local. And I'll tell you why. If I go to cloud, you can see that it says this runs in the cloud even when the device is offline, which is great. chat project

9:54 model and reasoning settings though are not available and that isn't great because specifically we want to trade with GBT6 Astra. So I'm going to go back to this device and now I can actually

10:02 choose the project to work inside and I can choose the model that we want as well as the reasoning. So I'm going to leave this on GBD6 Astra and I'm going to leave this on high. Now the other

10:10 thing you could do here is you could actually have this run in the same conversation thread every single time. And we're actually going to do that because that helps us set up the ability

10:18 to have you know the conversation history be stored. And Codex has this really nice autocompaction where I hardly even feel like I'm in context rot territory. So we're going to keep all of

10:25 these running together. So what I need to do is spin up a new thread which is going to be the trading challenge because we're going to have to set up what is it six different routines. And

10:33 then we're just going to point all six of these routines at the exact same chat to actually run in. And the cool thing is we can actually get all this set up with Astra and just using our natural

10:42 language instead of manually coming in here each time. So what I'm going to do is we're going to discard that routine and I'm actually going to start up a new project here. We're going to keep this

10:49 local for now. I'm going to open up the actual folder. But first, I'm just going to call this trading challenge. And then we're going to open this up to the one inside of my Herk 2 where we set up all

11:01 of those like files and things. So now I'm in the Asha trading bot challenge. I'm choosing that folder and I'm creating that project right here. And I'm just going to say, hey, I'm just

11:09 kicking off this thread. This is where all of our actual scheduled tasks are going to run in this one thread so that we can keep everything consistent. So just kicking that off so that the

11:16 routine can actually pull from it. And I'm just going to go ahead and rename this challenge thread. Okay, cool. So now I'm going to go back to the actual chat where I was

11:26 developing the strategy and I'm going to send off the SLGO prompt to help us set up these six routines. Hey there, Codeex. It's time for us to now set up the routines now that the project and

11:36 the strategy has been configured and you're connected to Alpaca. So you should be able to understand exactly what to do already because we talked about the strategy. I want you to set up

11:44 the six different scheduled tasks inside of Codeex. They can run locally on this device and they will each be their own individual scheduled task. As you can see, we've got these six different ones.

11:54 But what's important to me is that I kicked off a thread in a new project called trading challenge. And I kicked off a thread and I named it challenge thread. And I want all of these six

12:04 scheduled tasks to actually happen inside of this thread. So you're going to have to choose the chat when you're setting up each of these scheduled tasks. And that way we keep everything

12:13 consistent and um you know we have the continuity in that way as well as the actual progress logging and everything that we did like that with the handoff messages. We can just keep everything

12:22 more consistent and um I guess you know unified by doing that. So do you have any questions about this? Um let me know if there's anything that's unclear and ask me any questions and then we'll go

12:32 ahead and get everything set up. And by the way guys, there is a way to be able to set up cloud routines that can work with Astra. um just not something that I was going to dive into in this video.

12:40 But you can see now it understood what we wanted. It found the challenge thread and it's using that for all of the scheduled tasks. It will return to that same thread. So now we have these being

12:50 set up right now. And I could basically click into here to see how they are set up. So I can see this is the scheduled slot number one. It's giving it all this information. It's pointing to the right

12:58 folders and files that it needs to look at. It's telling it what to do. If we go to the end, we should see follow the shared startup and end of run handoff. save what changed, actual outcomes,

13:07 remaining risk, unresolved issues, and the exact next job, even on a no trade or failed run. We can see that this one runs in the challenge thread, which is exactly what we wanted. So, this is all

13:16 looking very good. Now, there is one more thing that I recommend setting up that I didn't yet instruct it to do, which is that I obviously want to be notified if something fails, but I also

13:24 want to be notified maybe at the end of each day or maybe a couple times each day, and I would like for that notification to be sent to ClickUp. So, that's one more thing that I'm going to

13:32 shoot off while this is actually working. And I can click steer to inject that message in. So I would like to have two updates per day. I would like one sort of like midway maybe near like the

13:42 afternoon check and then I would also like one after market closes on you know what happened that day and our current position and everything like that. So if you could just set up that automation.

13:53 So it sounds like we'll need one more scheduled task or maybe two. They should run in the same thread. But what I want them to do is notify me in ClickUp. So I want them to send me a DM. um into the

14:04 internal automations channel to ClickUp. And so this is where you can see I when I turned off this previous one, this one was sending me updates over here in the ClickUp channel and it should be doing

14:13 the exact same thing. And that way I can just get notified right here. Now I can see that the challenge thread is spinning. So I wonder if it injected a message over here. You can see this said

14:21 by chat GBT from another task. It said the six requested Astra schedules have been created for this exact task and local trading challenge checkout. They have temporarily paused during

14:30 verification blah blah blah. But this is basically what it will look like when the routine fires is we'll see a message come in and it will say sent by chatbt from another task and then we can

14:38 basically have full visibility as far as every time messages got sent, what happened, what Astra thought about. And that's going to be pretty cool for us to just have that visibility here. And of

14:46 course, we'll be able to have full visibility inside of our actual account as well because we can see all the trades and all the orders and positions and everything like that, too. You can

14:53 see that I just got a test message here sent to my ClickUp channel. Now, this got sent as me. It was my profile. If I wanted to log into a different profile like the UPAI one so that I got these

15:02 notifications in a different way, then I could do so. But for now, this is going to be good enough for me to be able to get notified when something has happened. All right, so I went ahead and

15:10 set up two extra scheduled tasks there. 1 p.m. and 3:15 p.m. Central for notifications, and it moved the midday report a little bit. Reports send even when there are no trades. Test delivery

15:19 succeeded. So everything is perfect. Everything now runs inside of the challenge thread. So everything will be right in here, which is going to be great. And you can see that there's a

15:26 little clock symbol next to this thread because it knows how all of these routines, if I go to the scheduled and we go to these, you can see that these are all running and they all start

15:36 tomorrow, which is why they're currently paused, but tomorrow they'll all be here and active. And then we can look at them. We can see previous runs and we can open up the chat, which once again

15:44 takes us to the spot that they're all going to be talking in. Now, what else is really cool is right here, I'm on my phone. So, this is my phone just being mirrored onto the desktop. And if I go

15:52 over here and I click on remote, what this allows you to do is essentially remote into your local desktop. This is connected to right here. So in the challenge thread, or actually let's not

16:01 do the challenge thread. Let's just do the develop 7-day trading strategy. And I open this up here, I can basically have this be perfectly synced, which means if I come in my phone and I just

16:10 say hi, that will shoot off in my actual desktop right here as well. So it's very similar to the Grockpot way or the cloud code remote. It's perfectly synced, which means even when I'm not home, I

16:22 could come into here and open up the challenge thread and I can always check in on what's going on with my agents here with the trading strategy. So, if you didn't know about the remote, then

16:30 definitely make sure you're aware of remote. But that really is going to do it for this one. You guys saw exactly how I got this set up. You saw the strategy that we're using here. And

16:36 obviously, I'm going to be watching this thing over the seven days to make sure nothing is going too wrong. And if you want to follow along and you're interested to see how that actually

16:42 plays out with Astro, then definitely stay tuned. That video will be coming out in a few weeks. And if it all goes well, I'll keep running it for multiple months because I think it's really

16:49 interesting to see how intelligent this stuff is truly getting and how it's able to learn from its mistakes in the past and change the strategy over time. But like I said, that's going to do it. So

16:57 if you guys enjoyed, you learned something new, please give it a like. It helps me out a ton. And as always, I appreciate you guys making it to the end of the video and I'll see you all in the

17:03 next one. Thanks everyone.