The free $0mo AI coding stack

Coding with AI doesn't have to be expensive. You can code with AI models for completely free, zero dollars a month. And this video is going to show you exactly how. This video is for people who either can't or don't want to pay for frontier models. I'm going to show you the best options you've got for coding with free models. And look, I'm going to be honest over here, while this is going to work pretty effectively for most use cases, it does have trade-offs. There's a reason the frontier models are the frontier. And so you can absolutely code with free models. There are going to be a few trade-offs and I'm going to go over the trade-offs as well in this video. And I want you guys to have realistic expectations about this video. You're not going to start solving Navier-Stokes math problems just by using some DeepSeek Flash model. But you can absolutely code some pretty amazing apps with these open and free models. You just have to, you know, use them properly. And so with that being said, let's dive into the free $0 a month AI coding stack. So I've divided this AI coding stack into two layers, basically two groups. There's web LLMs and there's coding agents. The web LLMs are basically the, you know, website versions of ChatGPT and Cloud and all these other models. And they're mostly here for planning purposes, planning, architecture, research and stuff like that. The coding agents, these are the ones that actually you install on your PC and they're going to do the actual coding for you. These are a bunch of harnesses that offer free models that you can use for coding. Now like I mentioned before, these free models, they're pretty weak, but you make up for it with better context and better planning. And that's where the web LLMs come in. We're going to talk about how to use both of these together. I want you to start using these web LLMs for things like general planning, for research, for just fleshing out your project and also for one-off scripts and code that does not depend on the rest of your project places where you don't need to send in you know other scripts or other code in your code base as context when you want to generate a one-off script for something use these web llms they have got pretty generous limits and you can generate a lot of code with them you can also do planning and other stuff regarding your projects and code base and they're really useful for that it's just that it's much harder to feed in context into these web LLMs because they can't read files on your computer, they can't read the entire code base. And so these are gonna be limited to planning and research and architecting your code base and generating one-off scripts. Use them as much as you can for those purposes so you don run out of limits over here in the free options that you got And also regarding context I want you to start using projects in every single web LLM provider you using ChatGPT supports them Cloud supports them. I'm pretty sure all the other Chinese options support them as well. Projects basically are like a container for all the context that you have regarding some project or something you're working on, right? You can create a new project. You can upload context like files and markdowns and documents and images and all the other stuff and this way you won't have to constantly feed in the exact same context again and again into cloud or chat gpd when you're using them for planning and architecting your code base you can have a bunch of markdown files that explain how your code base is structured what kind of project you're building you can put it all over here and then you can ask questions about it and help you know use these models to help you refine your prompts and to write better code in your code base. So use projects, make one for every single thing you're building on your machine, and just upload all of its context over here. It's going to save you a lot of time because you won't have to repetitively, like, repeat the same context in every single prompt. And the reason we're using projects over here is because web LLMs just make it so much harder to feed in context about your code base, right? If you were using a coding agent, you know, it can already read all the files on your machine. It can already run commands and stuff. It can get all the context it needs by itself. You don't need to do anything about that. But when it comes to these websites, they can't read any files on your machine. They can't get the context they need. And so you have to feed in that context every single time. And projects is how you, you know, reduce that effort every single time you want to send a prompt. And the reason we're even using these web LLMs in the first place is because they have much more intelligent models than all of the free options that you get in coding harnesses. Like for example, in Claude, you can get a model like Sonnet 5.5 for free. And this is way better than any Chinese or open weight or free model you could use, which means it's a great model to use for planning and architecting your code base and for writing one-off scripts, maybe even to do code review on some scripts you upload over here. It's an extremely good model for that. And none of the options you see over here are going to have a model as capable as that for free. That's why we use web LLMs for planning and architecting and other stuff like that. And we only use the coding agent to actually write the code. And so now let's talk about the actual coding agents over here. There's KiloCode. This is really great. I've used their free tier and you can basically just keep using it all day long without running into any limits The official limit that I found on their website is like 200 requests an hour But I be honest I used these models like all day long in like really you know, repetitive tasks and stuff. And I've launched multiple sub agents as well. And I have never hit the 200 requests per hour limit. So kilo code is probably one of the best options you've got. It's just that their models are pretty dumb and not intelligent at all. So you have to steer them very, very carefully. The second option is OpenCode, and this is becoming way better nowadays. At the time of recording this video, OpenCode has just made models like DeepSeek 4.1 Flash and GLM 5.3 Flash completely free. Like this just happened a couple days ago. I saw the tweet on X. Both of these are extremely capable models. I've used GLM 5.3 Flash a lot in my projects, and it is extremely good. Like for most coding tasks, GLM 5.3 Flash will get you a long way ahead. I haven't used DeepSeek 4.1, but honestly, I've only heard good things about it, and I'm sure it's just as good as GLM 5.3 Flash, if not even better. And both of these are available completely free in open code. So go ahead and install both of these harnesses. And along with that, you can use anti-gravity. There is a free starter quota available, but honestly, it's so small that you'll run out of it in like 10 tasks or so. Cursor also gives a free tier, but again, you'll run out of it pretty fast. Now GitHub Copilot has the more generous free tier out of these three. And so go ahead and use that as well to get some work done on your machine. But just know you'll probably run into rate limits over there as well. So use these three whenever you can, whenever you've got the rate limit for it. But you'll mostly be using KiloCode and OpenCode, since those are the most generous coding harnesses available right now. Now let's talk about how to actually work with these cheap and free models because the workflow that you use with all of these open and free models is going to be way different and way more painful than something like Opus 5.5 or something. I mean with Opus 5.5 over here, I could just give it a like a two sentence prompt over here and just let it run and it can do so many tasks for me without me having to specify it. But when it comes to these cheap and free models that are in these harnesses that you get for free, you're going to have to babysit them a lot. You're going to have to steer these models a lot. A lot of times they not going to do exactly what you ask them to A lot of times they going to forget to do certain things The code quality of these models isn going to be top tier And so you really have to give them very very specific very very precise prompts and you should not rely on their judgment at all Do not ask any of these free models what you should do Figure out what you should do and ask these models to do it Get it Because these free models, they're really good at coding. They're just not good at judgment and they're not good at planning and stuff. So you always have to give some really specific prompts, really precise prompts to get these models to do what you want them to do. They're not as good as the frontier models. And the easiest way to give those specific prompts is to plan your workout in a really smart model like sonnet 5.5 with all of your project context already added in like we saw over here in the projects tab right like i said before create a project for every single thing you're working on add all the context and then plan in a really smart model like sonnet 5.5 and ask it to give you you know a bunch of like prompts that you can feed into your coding agent that are going to give you the code for you to you know integrate into your project so let me switch back to Sonnet because I don't want to run out of Opus limits right now. But yeah, that is that is the stack that you're going to use to create projects in cloud or chat GPD or something. You feed in all of your context over there so you don't have to repeat yourself constantly. And then you plan in those really smart models and ask them to give you coding prompts that you can feed into your coding agent. And then you use those prompts to write the code itself. And that is how you can use cheap and free models to code pretty much anything you want. Now, I do have to give a bit of a reality check over here this is long overdue um if you can't pay for the frontier models like you're gonna notice a pretty big difference in what's available out there and all these free models that you're using and i hope i've been honest about the trade-offs between both of these in this video because like when it comes to these free models you're gonna have to babysit them a lot they're not as smart as a frontier model and you're gonna have to be really precise with your edits you're gonna have to steer these models a lot more and you cannot let them run autonomously on your machine with no approval layers from you so if you can afford to do it i'd highly recommend just saving up like 20 bucks or something buying the cloud pro plan or buying chat jibbitdy because the frontier models are extremely capable now and if you can't do that look i'm not going to judge someone for their financial situations or comment on your finances or or something. If you can't do that, these free models are going to give you a really good shot at building any app that you want. It's just going to be a little more effort on your part, right? And I hope that all the options and the guidance that I provided in this video helps you to build whatever it is you want to build. That's all for this one. Like, comment, subscribe. Thanks for watching, and I'll see you in the next one.