How to Use Claude Code for FREE (2026)

Hey, this is a video on how to use Cloud Code completely free.

So, Cloud Code is the number one coding agent in the world right now, but it also costs between 20 to 200 bucks a month. It is pretty restrictive rate limits as we know, and some people complain about quality fluctuations. In this video, I'm going to show you a free fully open-source alternative method called free Cloud Code that literally lets you use the exact same Cloud Code interface. So, the same terminal and the same commands, but with significantly more affordable back-end models that allow you to achieve somewhere between 80 to 90% of the quality of Opus 4.7, but for between 2 to 5% of the cost. So very high leverage. Let me show you how to do it.

So this is an app called Habitual that I just put together and uh it's just a simple habit tracker app. And what's funny is this would have cost me something like $5 to $10 in in Anthropic credits with DeepSeek Flash V4, which is the model that I was using inside of this free Cloud Code uh configuration. I was able to build this for like $3. So that's not hyperbole. I literally did so for like several hundred times less money than I would pay to Anthropic. And it's not to say that this is the exact same quality of app that I would get if I built, you know, using Cloud, but uh, you know, the the coding models that are available now on platforms like OpenRouter, NVIDIA NIM, and then Ollama, all three sources which I'm going to run you guys through how to set up on your end are actually quite powerful and you can do more or less anything you would do with Cloud Code um, you know, with one of those models instead of a traditional Opus back-end.

The thing to know is, you know, obviously at really high levels of quality, which I think like the Cloud Opus models are getting to, even a 1% improvement in quality might mean like really, really different results. So I'm cognizant of that. But I think a universe probably exists where you can fire off a lot of things with like Opus 4.7 or another really advanced uh kind of frontier model and then have, you know, DeepSeek, let's say, or GLM 4.7 do a lot of like the heavy lifting work behind the scenes, the code refactoring and so on and so forth uh to save you like 5 to 10x what you what you normally would have spent.

So what I have over here on the left is I I actually have DeepSeek running. You can see it's doing this in the exact same Cloud Code interface as per usual. We even have like thinking blocks that are kind of coming out. Um, you can actually see that it's referencing DeepSeek V4 Flash via proxy. And that's going to be the approach I'm going to show you. But um, for all intents and purposes, I mean, not really much else to say. You talk to it the same way that you do any sort of like Cloud Opus model. Let me show you guys how to actually set this up. It's super easy. Takes just a few minutes. And then I can show you guys some examples of different models and how to how to work with them.

So the way that typically Cloud Code works is you have it set up here in a terminal. Maybe you have it in Anti-gravity, maybe you have it in like the Cloud Desktop app or something like that. And then you send it something. So you say, "Hey," and you know, the second that you send it a request, what it'll do is it'll push that over to its API right over here. And then you'll be billed usually based off token usage directly or, you know, if you're smart, what you'll do is you get a plan like um Pro or maybe Max or whatever. And what these do is they basically subsidize the token usage a little bit in exchange for Claude or Anthropic, the parent company, knowing that you're going to pay a subscription every single month. And that's all well and good. It's just this is a fair amount of money, right? I would much prefer to spend $0.0005 per million tokens instead of $5 per million tokens.

Now, what our solution is going to do is when I send something over to Cloud Code, instead of it just going directly to Anthropic, what we're going to do is we're going to force it to instead push to our free Cloud Code proxy. And uh this is going to be set up as a local server on our computer. Hence why it says localhost, you know, 8082. You can put it on whatever freaking server you want. But what this proxy will do then is it's then going to route it to one of three possibilities. The first is NVIDIA's NIM, which uh is free, you know, with an account. I think you can also pay for some that basically leverages a bunch of NVIDIA GPUs to send requests. These models aren't as good as this next one, which costs some money. Uh there's OpenRouter, which uh, you know, you can get models down to like 14 cents per million. That's quite a far cry away from $25 per million, right? And then finally, you have Ollama. And Ollama actually uses like your your own GPU. So you might have a GPU rig set up or I don't know, maybe you just have a gaming laptop or something. You can um set up models to run directly on that infrastructure. And if you're really good, you can actually set them up to run way faster than traditional uh, you know, Cloud models cuz you're not competing with millions of other people to send those requests.

So what we're going to do right now is I'm going to show you guys how to set up this section, which is the free Cloud Code proxy. And then after I'm actually just going to run through um all three of these options as well as, you know, all of the models sort of on these platforms so you guys could see everything that you need to do in order to set them up yourself. And you don't need any technical experience setting up local models or anything like that. I'm going to I'm going to walk you guys through it and handhold you from start to finish.

Okay. So everything starts with this repo right here called free-cloud-code by Ali Sharer. And it looks like this has gained a ton of popularity in a very short period of time. Just scrolling down here to its star history. It was basically at nothing in February and March. And just now it's hit an inflection point. A lot of people have been sharing it and stuff like that. I want you guys to know it's not about the specific repo. There are a lot of very similar approaches that do basically exactly what this thing does. So, uh, should you not want to use this one or want to use a different one or should this become unavailable at any point in time like just, you know, don't the point the purpose of this guide is not to get you hooked on this one particular solution. The purpose of this guide is so that you guys just see it in practice and then know how to set it up in your end.

Okay. So, what we need is we just need this little quick start down here. And I'll walk you guys through everything. Um, just open up a terminal. And so this terminal instance is currently running, you know, DeepSeek V4 Flash. This is like the one that's actually live. Uh, so what I'm going to do is I'm just going to exit out of this. And then I'm just going to clear. Okay. If you don't know how to do that, let's say you have another terminal window open. You can also just head over to shell new tab. If you're running uh Windows, then you're going to have to use like Windows specific thing called PowerShell, which is a little bit different. But um, this quick start guide is going to run you guys through how to do everything regardless.

So because I'm on Mac OS or Linux, all I'm going to do is I'm just going to copy these three commands in. It's now going to go through and then install everything that it needs to. In my case, I already have all of it installed because I've obviously set this up before. Okay. Then what I'm going to do is I'll just copy this next command over. Give that a paste. So, it's going to clone directly in. Now, uh, when I say clone directly in, you'll see this has actually changed. And that's because one of the three commands here was to go into the directory. And so, now we're actually in the directory. If you're on PowerShell, you just copy this command over. Um, same thing with that. Okay.

And so, from here on out, we actually have like the folder set up. Like I know it sounds interesting and weird, but we've we've done like 80% of the work. Um, just by copying this configuration, we have more or less everything we need aside from just the API key of the provider that we're using. The simplest and easiest provider to get set up with is OpenRouter. So that's what I'm going to show you guys first. And then after that, we'll chat NVIDIA NIM and then finally Ollama. OpenRouter is probably the simplest like plug-and-play drop-in replacement for whatever you guys are currently using. If you guys have any experience sending API requests or whatever before uh this video, then that'll be very, very simple to you guys. And then if not, you know, I'll show you guys how to use even cheaper ones.

But basically, what you need to do is you just need to go and grab your your your uh OpenRouter uh uh path. So that's pretty easy. All you need to do is just go to openrouter.ai. Now, I actually already have an account set up. So I'm just going to open up my account. And then uh here are all my requests that I've been making. I'm just going to go to API keys and then I'm just going to go create. I'll say for YouTube. Then I don't know. I'm just gonna expire this in like a day. That way if I forgot the API key, nobody can uh, you know, bill or run me up a million dollars. I'm also going to copy it. And you have to copy it right there. Otherwise, it's going to uh basically exit out.

And then what I'm going to do, and uh this may be kind of intimidating if you've never done it before, but I'm just going to find that folder. So free-cloud-code. And the way that you can do this really quickly here is just `pwd`. So in my case, `users/next-world-cloud/free-cloud-code`. That's pretty straightforward. So I'm going to go to Finder. And uh I already had this open earlier. So I'm going to go hello-world-cloud. And once you're done with that, you'll see the folder. Um, what you guys will probably notice right off the bat if you just have like a vanilla Mac install on our developers is these files here are sort of like um, you know, significantly less opaque. They're sort of shadowed out and you may have never seen files like that before. What these are, these are hidden files. Anything on a Mac with a period in front of the file name basically just defaults to hidden. This is a convention that they use just so that you know if developers build something cool and and maybe potentially a little dangerous. Um, average, you know, computer users don't really have to worry about that. Um, but what's really cool is you can actually just show or hide hidden files at any time. And I've totally forgotten the hotkey. So I'm just going to look up like Mac OS hidden file hotkey. And uh right over here you can see it's `Command + Shift + .`. So I'm going to go back over here, hold `Command + Shift`, press `.`, and boom. It just like removed all the hidden files and then it's showing them. It's removing them. It's showing them. So yeah, if you guys ever wonder about how to do that, um, just feel free to to look that up and it'll walk you through it.

And the file you want to edit here is this `.env` file right here. Okay. As you see, if I zoom way in on this, uh, once you open this up, there'll be placeholders for all of the API keys. OpenRouter config, DeepSeek config, you know, LM Studio config, and so on and so forth. Okay. So I'm going to paste the key in right over here. And then there's one more thing that you have to do. If you scroll all the way down to where it says OpenRouter, you also have to specify a model. And uh, you might be wondering where do I do that? If you scroll down, there's a little section where it says `model`. Okay, right over here. What you have to do is you have to you have to choose the provider from OpenRouter. So basically, if we want to use um DeepSeek 4 or something like that, we just go to browse all models and then I'm just going to type `deepseek v4 flash`. This one right over here. There will be this little copy right over here. `deepseek/deepseek-v4-flash`. Okay. Once you pasted that in, go `open_router` like this. And this should be the format. Um, I've done this a couple of times. You see here it's `openrouter`. That's sort of the provider. Then `deepseek/deepseek-v4-flash`. If not, we'll figure it out. Okay. I'm going to save this.

And once I'm done with that, um, I can actually now open up a terminal right over here. Verify that I'm inside of this directory. Okay. And we just did the editing on the `.env` file, which is hidden. So, you need to go `ls -a` if you want to see it. And then from here, we can actually just scroll up to step three, which is start the proxy. Paste this in. And I may have this port set up. No, I don't. Which is nice. Cool. And now this is running. And now we just need to open up another terminal window. So, a lot of terminal windows here, but let's just open that up. Let's just zoom way in like I did a moment ago. I think that's about equal, right? This one's a little bigger. And then uh we just run Cloud Code with this proxy setup. Okay. So, I'm going to copy this bash command, paste this in, and then go enter. And you can actually see we're running `deepseek-v4-flash` via proxy down here because we just made that request and we sort of did that specification. And I'm going to go `hello`.

So when you say `hello`, basically this right over here is just a real-time log where it's catching every incoming request. It's then sending that incoming request to, you know, DeepSeek or some other provider. DeepSeek is then catching that and then they're sending it back. And so, you know, on the first initialization, it may take a little while and that's more or less what happened here. The user just said, "Hello, simple greeting. I should respond in kind with a concise, friendly greeting." "Hello, how can I help you today?"

Now, what's really interesting is when you ask it, "What model are you?" Because there's so much Cloud stuff um sort of baked into the the system prompt, it doesn't always know. And so, it's saying it's Cloud Opus 4.6 with the 1M context. So, when running is Cloud Code, the CLI agentic interface. But what's really interesting is despite the fact that it's really confident that it is, you know, if you go back to your OpenRouter, uh, sorry, you got to go down to logs, you can actually see the request that we just sent. It's 539. I just sent a request over here to DeepSeek V4 Flash. And uh, you know, you can actually see the the raw JSON down here with said request as well. Like this is the same request that we just sent a moment ago. And what's cool is, I mean, this one here uh, let's see, the input token was like 30,000 or so. That's the whole system prompt of Cloud Code. And then the output was 47, which is literally like, "Hello, how can I help you today?" Uh, so we do unfortunately have to feed in the whole system prompt of Cloud Code. But I mean, that is literally what Cloud Code is, right? It's just part of the deal. But yeah, now you literally have like another model set up here. We're using DeepSeek V4.

Awesome. Build me a simple habit tracking app in a directory. Let's call subdirectory called `habit-tracker`. Make it super straightforward. Local. This is a demo. And I just want to give you guys like an actual live play-by-play because, you know, I think local models, well, not local models, sorry. Non-frontier models. So like DeepSeek instead of, you know, one of the big players, Anthropic, Google, you know, OpenAI, you know, like it's not as good and it's probably not going to be as fast. You can actually see the number of tokens that it produces over a time period. Um, you know, like represented down over here and you'll see the speeds vary from time to time. This one over here was 30, 23.6. This one was 59.4. So like, you know, it actually does take a little longer to do some things. Um, counterintuitively, it takes a lot less time to do other things because it's like different architecture, right? These models uh and you know, DeepSeek in particular, I think is is out of China. These ones are like built with very different architecture in mind and they work a little bit differently. So it's just worth it to like actually see what's going on and that's what I wanted to do here.

What's really cool though is the whole Cloud interface works basically the exact same. You see here how we have like the transfiguring. We got the cute little messages. Uh, it's clearly doing some thinking. Uh, you know, we can press `Option + Enter` to send a multi-line message. We can do a lot here. And now you can see that we actually have the app. So the app lives at `nyxive/habit-tracker`. So, what I'm actually going to do is I'll just say, "Open it in Chrome," and let's see if it can uh it can get that. So, just said, "Open it in Chrome," and then boom, we have it right over here. Hey, what's up? So, I'm going to add it. And now I'm just going to kind of check my little habits. Boom. Just did it. Fantastic. I did all seven.

Because of this, there are other strategies that you should use as well. Like, for instance, a common strategy that I like using is I just create a new instance every 50,000 tokens or so because I find that quality goes down every 50,000 or so tokens. Uh, that, you know, it's just not as smart or as capable. Obviously, you're trading off that intelligence and capability for money. So, what I did here is I just opened up a new uh, you know, instance and then I said, "Hey, here's a file. This is way too simple. Make it look lux high quality. Add a serif font. Make it a much higher brow feel." Uh, and now it's running. So, as it correctly identified, the JavaScript is pretty clean there. So, what it's going to do instead is just redesign both the HTML and the CSS, which is pretty neat. You can actually see the updates occurring basically live. I'm just refreshing this over and over and over again. Uh, you know, it's already correctly established a nicer and sexier font, which is nice. So, I don't know. I mean, this isn't like an actual habit, right? So, what am I going to put here? I'll say, "20 push-ups," you know? And here I have my little thing. Nice colors, too. And, you know, after another five or 10 minutes, I'm sure you could see how you could build something like this just as a demo. I should note that I did this entire build in like entirely using DeepSeek. Whereas as I mentioned, I think probably the best thing to do if you really wanted a good combination of both cost and effectiveness would be to use a smarter model as an orchestrator and then have it send requests to smaller DeepSeek models with more limited context like let's say 50 or 100K. Um, just to ensure that like everybody's doing what everybody knows how to do the best.

Okay, so that was OpenRouter. Let me show you guys now how to do NVIDIA NIM. It's pretty straightforward. Same idea. Just scroll down here to where you actually have your uh NVIDIA NIM API key section, model section. And then what you have to do is I actually have to go sign up to NVIDIA's NIM platform. Now, in my case, I think I'm probably already signed up. May it may already have me signed in here. So, let me see. Yeah, it did. Um, what I'll do just to show you guys sort of what that process looks like is just sign out. Okay. Now, all I have to do is sign in with my email address, which I should be able to do here. The signup flow is probably going to ask you to put your phone number in, your email address, all that stuff. Um, I'm just going to assume that you guys have phone numbers and are willing to do that. I also just don't want to share mine. And then from here, just click generate API key. And then I'll just say for YouTube. Okay. And then the expiration here. Again, I'm just going to do I don't know, why don't we do 12 hours. Then click generate key. And now you have the actual API key right here, which is awesome.

Next thing you got to do is go to models really quickly. Um, when you click on models, assuming that NVIDIA actually loads, holy moly, that's slow. Um, anyway, it'll open up this little models page and then you can actually select whatever specific one you want. So, as you can see, you know, DeepSeek is over here. So is GLM 5.1, GLM 4.7 right now is free, so it doesn't even cost you any money. Um, that's what I'm going to be using just because I think that's probably like the simplest one to get set up with. And then, um, just because this is taking a while, I'm just going to go `Z-AI/GLM-4.7`. Okay. Now we're actually in a little test page and we can actually have a conversation with it. We could actually say, "Hey," um, this is now going to send the request to the model. The model's going to go through its whole thinking uh whatever and then, you know, so I guess we just verify that that works. Not super relevant for us. All we have to do now is if you go back to this quick start, you'll see that it says `nvidia-nim/z-ai/glm-4.7`. So I think we actually probably already have this set up. I'm just going to double check. Just go back to Finder where you had that `.env`. Okay. Then from here, um, just scroll down until you have the NVIDIA API key, which we do. So I'm just going to paste in my NVIDIA API key. Zoom in a bit for you guys. Then scroll scroll down to again where it says `model`. And this is going to be where uh you'll put in the NVIDIA one, the one that we just had a moment ago. So I think that's probably the one that we'll do. I'll use NVIDIA NIM's GLM 4.7 because it's free and, you know, I want to show you guys how to do a free model. We'll paste that in. And now I'm just going to hold `Command` and then press `S`. That's going to save it. Uh, you can you can see there are a lot of other, you know, different types of models and so on and so forth.

And now what we can do, you know, this is currently actually doing a build right now. Um, but as you can see, the tokens have stalled. And the reason why the tokens have stalled is because I just like changed the the file. What I what I'm going to do is I'm just going to delete both of these. Okay. And I'm doing that by holding `Control` and then pressing `C`. I'm going to clear these out. And now I'm just going to run the server again. That's 8082. And then over here I'm just going to `anthropic-off-token`. `anthropic-base-url=localhost:8082/v1`. Okay. So now I'm going to go, "Hey." It looks like the reason we're getting this is speed. I think that's because I have fast mode on. So, let me just turn that off. Okay, cool. So, we just went, "Hey," and then it's sending a request over to NVIDIA now. So, now it's saying, "Hey, how can I help you today?" Fantastic.

What you'll find is if you're using things that aren't supported by these models, like for instance, these models don't have fast mode, even though that's sort of like enabled within Cloud Code. You send the request, you'll receive an API error being like, "Hey, what the hell's the speed thing? I don't have that." So, I just make sure you turn any of that any and all of that stuff off. Uh, and then I'll say, "Hey, what's your working directory?" Just because I want to double check and see what it's at. And, uh, you'll also notice that this is still broken. It says `DeepSeek V4 Flash`, which is kind of funny. Uh, this one says, "Hello World Cloud, free Cloud Code." Awesome. Um, you know, how do we change the status line? Let's just ask it that. Maybe you can actually do that. So now you can see here that that response took a little while and then it asked about changing the status line. Um, this thinking has occurred here in just one giant block as opposed to streaming, which is interesting. So what it's doing now is it's spawning the status line setup agent, okay, which is something that's actually built into Cloud Code, interestingly enough, but again, we're using a different model brain to power it. And so that's what's really cool about this harness, like it's interoperable and you can swap whatever other agent sort of inside of this container. Okay. And here it's now asking me what the model name is. For model name, I'm just going to say, "Use GLM4.7 for model name." And then everything else the same. And let's see if it can actually go do it if it has enough context and intelligence to do it natively on its own. Okay.

And the last thing we need to do is we need to set up Ollama. So it's actually pretty straightforward. Just head over to ollama.com and then there'll actually be a one-liner that you can just paste into your terminal. You can also download it through like more conventional means. Um, but I'm just going to use the shell because that's easy for me. These guys are pretty cute. They're a lot more visual. So all your stuff will be really adorable. Once you're done, just type in `ollama`. It'll actually like get Ollama up and running on your computer. So as you can see here, there's now `chat with a model`, `launch open-webui`, `launch ollama-webui`, `launch open-code`, `more`, right? There's a lot more down here. Whatever files you want. Um, just for examples, I'll go `chat with a model`. And then what do we want? We want a really simple one. Uh, you can see a couple of these are Cloud, right? So this one's actually in the Cloud. That one's in the Cloud. That one's in the Cloud. That one's in the Cloud. But I want something like really, really, really basic and really simple. I think I'll probably just use Gemma 4 because this one's 12 gigs and I can probably download it through the video. So, I'll just click Gemma 4. Download Gemma 4. Now, it's going to pull the manifest. I actually have to do the the download. Now, my internet's pretty fast, so I should be able to download, you know, this whole thing, which says it's around 10 GB or so in a couple minutes. I'm just going to circle back when it's done. We'll also go back here to models. And then what we wanted was Oh jeez, what did we just download? I think it was Okay, awesome. So we actually have Gemma 4. And the reason why I'm doing this is because I just want to get the uh the code and it looks like the code is just `gemma-4` all lowercase. So we should be able to do that pretty straightforwardly when it's done. Okay. And then I think we're now actually talking to it. So I'll just say `hey`. And yeah, you can see the thinking process is now being written. Uh, what's cool is this is now entirely local on my computer. Um, are you on the Cloud or what? I don't actually know if it's going to know. It's asking a philosophical question. Yeah. No, it clearly doesn't actually know where it's running. Um, that's pretty funny. Anyway, regardless, uh, we have basically everything that we we need set up, which is cool. So, I'm just actually just going to exit out of this. I think it's `Control + D`. Okay. Once you're done with this, I'll just `Control + C`. Okay. And I'll clear it. Okay. And then once we're done, um, we just need to write `pull` and I'll pull `gemma-4`. That will actually like go and grab Gemma 4. I mean, in my case, I already have it on my computer because we just downloaded it. But from there, we just go `serve`. Uh, and I think the reason why it's served is it I I think it like auto-serves. Um, when you do that now, it might not have used to do that. Alternatively, we might have just already been running it, which is great. Um, that means we can now go through and then update the last thing that needs updating. And then open up the file and then look for Ollama. You can see down here it says `config.local.provider=openai`, `api_key=`, `base_url=http://localhost:11434/v1`. And if you guys notice the base URL, `localhost:11434` is the same as the `11434` here. And the `localhost`. Another way to say that is just `127.0.0.1`. So what this is saying is like, hey, you're currently running the server for Ollama on this, which is obviously awesome to us. Okay. And then the last thing we have to do is we have to actually grab um whatever this is `gemma-4:latest`. Okay. And then copy that over and then paste it in right here. And we always have to prepend with `ollama/` if it's if it's listed up here. So `ollama/gemma-4:latest`. Save that.

And then now we can actually go and run the app again. So we just ran the app and then uh I think it's over here. So we connect to that and now we should be having a conversation with um Gemma 4. So I'll say `hey`. Let's see how it goes. You can see this is sent, I think to a different endpoint. So this tells me it's probably right. What's really interesting is I just sent that and my computer just started getting really hot. The reason why is because all of this was occurring locally on my on my own computer. I can actually hear it humming now, which is really funny. Um, that's because it's literally passing through that request and doing all the transformer multiplication, uh, matrix, you know, inversions and whatever the hell, like literally on my own computer, which is really cool to me. And it's now dying down because it's not actually doing the the run anymore. So, that took a fair amount of time just to get this response, right? And the reason why is because when you run on local models like Ollama, um, it's it's really slow, like to be honest, because it's running on your own hardware, right? Uh, it's different if you're running on like dedicated GPUs like with OpenRouter or um, even like NVIDIA NIM. But here like I am literally running on my MacBook and my MacBook is not optimized for this at all. So hopefully I'm still capable of recording without uh skipping frames or whatever. But yeah, that's uh that's how you run it locally on your on your computer as well. And I should note that this is a pretty big model. You know, it's like uh, you know, 10 GB with 128k context running locally. You know, if you don't have a pretty good computer, you'll probably have to use a dumber model. So um, I don't know. You can go newest tier to sort by like the the good ones. But a pretty reliable one was the Llama series of models for a while. So, uh, you know, if you're running something a little bit smaller, maybe Llama, let me see, Llama 3.1, 3.2. These ones are okay. Um, this guy said it was Mythos, which is hilarious. You'll also see that models on Ollama are from different people. This is by Sadiki Siddiki Nashra, which is pretty interesting. And so, typically what they'll do is they'll like modify the model a little bit. They'll change it. They'll fine-tune it. They'll just make it make it different. So, feel free to take a look see through Ollama's um database of models and see if there's anything maybe that like works for you. A payroll specific model, accounting specific model, and what have you.

One more thing that I'll mention is um earlier I talked about having Claude orchestrate conversations between DeepSeek and then, you know, a smarter model in order to do stuff. And so, what I've done here, if I just get my ugly head out of the way, is um is I've done that. I've uh basically said, "Hey, send a message to DeepSeek using the flow we created earlier. Then tell me what it says back." And so that's what it's doing right now. It's actually basically doing all this hosting stuff totally on its own. And as you can see, when you just send a message directly to DeepSeek, it says, "I am DeepSeek, the latest version of the AI model developed by DeepSeek company." Um, you know, okay, great. Orchestrate it to build an app, guide it, and suggest it through the steps. App description. I don't know, simple calorie tracker, but MVP needs to appear super lux and high-end for the purposes of a demo. And um, you know, what's happening right now is like my much smarter model, Opus 4.6 in this case. I know it's saying GLM 4.7 down here, but that's because we actually just updated the status line manually, which you shouldn't probably do, by the way. I hardcoded in GLM 4.7 just as proof of concept. Uh, what we're doing is we're using this much smarter Opus model to basically communicate with DeepSeek through the proxy server that we just set up. So it's actually using, you know, Cloud Code under the hood and then we're going back and forth. And as you can see, our simple prompt here was understood, digested, and then spat out by Opus. So it's a lot more complex and actually has some opinionated baked-in font choices and colors and stuff like that. And um, it's just sending all of this over to DeepSeek right now and it's just going to go back and forth with DeepSeek until we have, you know, an okay-looking app. And as you guys can see, the orchestration sort of approach put together this app, which is kind of neat. Uh, this is just a a base calorie tracker, but still looks pretty cool. Obviously some minor issues here like the spacing. I think if I built in a better orchestration loop, like I had it do screenshots and stuff, it probably would have sorted that out. Even has some like quick add calorie stuff, which is wild. Then you actually have the the log here as well. So you can get, you know, arbitrarily complex, more or less as complex as you want. Looks like it even did a little calorie tracker export, which is awesome. So that's pretty cool. And um, you know, when Anthropic ran the same sort of infrastructure, uh, I think pairing Opus 4.6 with I think Sonnet 4.6 or maybe it was Opus 4.5 with Sonnet 4.5 sub-agents because this is basically what we're doing is sub-agent flow. Uh, I think it performed like I don't know, 15% better or something like that. So what we're doing now is we're not just using sub-agents made by the same company. We're using subs that are hundreds of times cheaper to do a lot of our intellectual heavy lifting. And then the Opus series is just going to take a look at it and be like, that's good or that's bad, which is obviously a lot more profitable.

Okay, so hopefully you guys appreciated that video. I had a lot of fun putting it together. Uh, costs were a major concern of mine up until just a few years ago when I started uh, you know, making a fair amount of money with the stuff. So obviously this sort of intelligence should be equally distributed, but for the near term, I doubt that it will be simply because, you know, uh, model companies got to eat. I guess GPUs got to go burr. If you guys like this sort of thing and you want to see how to maybe make some money as opposed to try and squeeze your wallet as much as you possibly can, definitely check out Maker School. It's my day-by-day accountability program where I basically show people how to take their skills with Cloud Code and then other no-code automation platforms as well like Zapier and so on and so forth and then use that to sell AI agents and full workflow automation to companies to make a living. Uh, we have a 90-day guarantee where you will get your first client within 90 days or I will pay you all of the money that you spent on my program back. Uh, it's been run through by over 10,000 people to date and I'm not going to sit here advertising any more than that, but it's pretty cool. There's basically no downside to signing up.

Aside from that, if you could do me a big solid, just like and subscribe to the video. Uh, every little bit of engagement helps and I'm starting a big YouTube run here, so that'd be quite valuable. And, uh, yeah, let me know what you guys want to see next because I make videos like this basically entirely based off of your requests and recommendations. This one was made directly by a request that I received a couple of days ago. Have a lovely rest of the day and I'll catch all y'all in the next.