Claude Code is the most powerful coding agent in the world. The problem is it costs a fortune. You pay every month and the second you hit your limit mid-task, it just stops.
You can either pay more or wait. So, here's the good news. You can now run Claude Code completely free on a model that lives on your own machine.
No, this is not a Claude Code knockoff. This is Claude Code but on a different AI model. So, in this video, I'm going to install [music] it, set it up, and show you exactly how to run it.
By the end, you'll have the same Claude Code you already use except [music] it will cost you nothing. So, let's get started. Here's how this works.
Claude Code is really two separate things in one. There's the AI model, aka the brain that does the thinking, and Claude Code that's Fable, Opus, Sonnet. That's the Anthropic model family.
And then, there's the tool wrapped around it, Claude Code. Some people call it the agent harness, which is just a fancy way of calling the app that wraps around the AI model. When you pay Anthropic, you're paying for the brain, not the tool.
The tool runs on whatever brain you point it at. So, all we do is keep Claude Code exactly as it is and point it at a brain that runs on your own computer instead, for free. And the thing that makes this possible is Ollama.
It runs AI models locally, right on your machine, and recently it started speaking Claude Code's own language. So, the two connect directly with nothing in between them. So, let me show you how to build it.
First, you install Ollama and give it a model to run. To do this, you go to the Ollama site and you click on get started. From there, you copy the install command for whichever operating system that you use.
I use a MacBook, so I'm going to go with Mac OS. I'll copy the install command and then open a where is it? Terminal window and paste that command in, hit enter and as you can see it will start downloading Ollama for your desktop.
That's it. Once it is done, you can either run it by typing in Ollama or as you can see, we now even have it as a normal application that we can just run and open from here. Then you download a local model.
Now, which local model you download depends on which computer you run. The more memory or RAM your computer has, the bigger and smarter model you can run. To know which model you should download, check how much RAM your computer has.
So if I go to my MacBook, I can see it has about 18 gigs of RAM. If you're on a normal laptop with around 8 to 16 gigs of RAM, then download Gemma 4 12b by just selecting it. If you've got 18 to 32 gigs, then download Qwen 3 coder, pick it and that's it.
And if you've got 64 gigs or more, then you can download GPT OSS 120, which is where the model will feel a lot more like the paid thing. Oh, and if you're on a Mac like I am, it's important you know that your memory is shared across everything. So if you're on a Mac, pick a model that's about half your memory size.
If you have 18 gigs of RAM like me, go for the models that work for half of that size. So 9 gigs models like Gemma 4 12b. And from there for that model to actually download, you can just send it hi and the first thing it will actually do is start downloading and installing it.
For me, I've already installed it, so I have access to it right away. Now, I know figuring out which model works best for your machine sounds like a headache. So, I made a free list that tells you exactly which model to run.
I've included it in the first link inside the description. Just find your RAM and it'll tell you what to download. If you're an AI founder and you're building something with AI, then make sure to subscribe to that same newsletter so you get every single drop completely free of charge.
But, when you download whichever model is best for you, you will now own that model for free on your computer. And now that we've downloaded our model, let me show you how you can point Claude Code at it. Pointing your local model at Claude Code is now easier than ever.
All you have to do is, if you have your Ollama app still open, click on apps and you'll see there are two ways to run Claude Code here. The first is by connecting it to the Claude Code desktop app so that it literally opens the Claude Code app for you with the model of your choice. And the second way is to click on Claude Code in other apps.
This will copy a command into your clipboard so that all you have to do is, again, go to your terminal, paste that command in, click launch or hit enter rather. And the command is literally Ollama launch Claude. Pick the model that we've installed, which for me is Gemma 4 12B, and it will start running and boom, just like that.
Look at that, Claude Code Gemma 4 12B with high effort. You now have the real Claude Code running on your local model. Congratulations.
And that's the whole connection. Claude Code now works exactly like it always has. It just costs you nothing now.
Now, let's give it a job and test it. I'm going to ask it to create a calculator app for us and open it. Now, I am filming this on my MacBook Pro, which runs 18 gigs of unified memory, which means I'm using a Gemma 4 12B for this tutorial.
And as you will see, given this is a very small local model, it will take its sweet time until it finishes this job. But that's what you need to understand about something being free. Your machine will have to do the heavy lifting instead.
And if you don't have a machine that can, then the trade-off is your time. Prompts take a lot longer to process. But would you look at that?
It finally finished, so let's test it out. 2 + 2 is 4, correct. * 8 is 32 / 8 is 4.
There we go. We have a calculator app. Running a model locally is the private way to do this, but it's not the only way to get Claude Code for free.
So, I thought I would show you two more ways real quick. The first one is a tool called Free LLM API. Instead of running a model on your machine, it stacks the free tiers almost every AI lab gives you.
Google, Grok, Mistral, and a bunch more behind one single endpoint, and it points Claude Code at it. You bring your own free keys, they stay encrypted on your machine, and you get way more model choice than you could ever run locally. The trade-off is that it's not private the way running local models is.
Your code still goes out to the providers, but if you care about free more than you care about private, this is probably the best option for you. This is the repo right here, and to install it, you first have to install Docker. Pick whichever version works on your operating system.
I'm on a Mac, so I would pick for Mac. Then, once you set this up and it's running, you open the Docker app, and you go ahead and open a terminal window inside of the Docker app, and then you paste in this command. As a reminder, all the commands, every single thing I'm using is in the first link in the side description below.
And once it is done installing, we grab the dashboard link, which is over here, and we're going to open it. Now, I'm already signed in, but I want to show you what to do if you're not. Since you're just going to be installing this, it's going to ask you for a setup code.
What you have to do is go back to the Docker and run Docker PS. Then after Docker PS, you're going to run this command, which if you notice this section right here, we're inputting with the name right here. So, those two should be the same.
If you press enter, it doesn't show me because I've already set it up, but it will give you the setup code, which you will use to input here, and boom, you're in. Then we go to keys, we go to add key, and we pick whatever free providers you want to add from in the API keys. I recommend you go with Google AI Studio.
There's Grok as well. There's Mistral. There are plenty, but if we go with Google AI Studio, we can click on get API key.
If you don't have an account, you can create one. You go to create API key, you name it whatever you want, say test one, create a key, and it will give us an API key. We will copy, we will paste it here, and we will label this whatever we want.
We will add the key, press save, and boom, just like that, we have that key. Then inside your terminal app, you paste in MPX free LMM API setup cloud. It will ask you for the unified API key, which you get from keys.
You go to unified API key, and over here, whoops, I almost showed my key. You copy, and you paste that in. Let me actually expand this window, and I'm happy that we got this error because if you get the same error, it's pointing at a different localhost than the one that we have.
Ours, for example, is 3001. It's pointing at localhost 3000. So, instead, you can paste this command instead, which points it at the correct dashboard.
Whichever dashboard you have, if yours ends with 3001, then put 3001. If it's different, then put that. Enter.
It's going to ask us again for that same key. We're going to go, copy it, go back in here, paste it, hit enter, and boom. Just like that.
You're done. There, you can run Claude in our terminal, and it will run using Perplexity's free tiers instead. The second option, and this is the one that I recommend a lot more because it is a lot easier than the one we just did, is for you to install Open Code.
It's a free, open-source coding agent built to run on any model you want from day one. No setup tricks needed. Luckily, installing Open Code is the easiest one here.
You just open Open Code, go to download, pick whichever you install, and you're here. With Open Code, you're not running on Claude Code. However, it is basically the same exact thing, minus all the setup headache [clears throat] that you just saw.
And they give you access to a bunch of free models from the get-go without having to go through any of that headache. Here, for example, in this test, I asked it the same thing, and it made us this app. Let me try to find it.
And there it is. It's literally the same exact thing, if anything, a little cleaner, in my opinion. And this was one-shotted.
So, that's the complete free Claude Code map. If you want full control and access to Claude Code for free, you run it through Ollama. If you want free Claude Code with more model choices, then you run it through free LLM API.
And if you'd rather just switch tools, then go with Open Code and skip all the installation steps. Pick whichever one fits you best. So, tell me in the comments which one are you going to try out first?
Oh, and as I mentioned in this video, the exact setup, every command I ran and used, and the models I'd actually pick are all listed for free in the first link in the description. If you enjoyed this video, then leave a like. Or if you're new to my channel, then subscribe because I've got a ton more content just like this coming your way.
Oh, and would you look at that? The algorithm gods have decided you're very likely to enjoy this video next, so click it, and I'll see you there.
