How to Use Codex + Claude for FREE!

Julian Goldie (0:00)

Today, I'm going to show you how to use Claude Code and Codex for free using a powerful system where you can basically plug in free local models as an extra model inside your existing setup. So that means you can use the power of ASTRA-6 or any different AI agent that you want, but at the same time, you can also have the option of running a local model for free or for private stuff, or for example, even running offline. That's another option. So how do we do this? Well, this is based on a question from Chad. So thank you very much, Chad, from the Air Profit Boardroom, who was asking, you know, I'm using free Claude code, which works.

I used Codex to tie it to my local coding model, and I would much rather use my local coding model inside Codex and Claude apps, but not fully replace the cloud models, just have the option in addition to, right? So basically that means, for example, you can still use ASTRA-6, or you can still use Fable 5.1, but at the same time, you probably want to switch between your local models too. So how can you do this? Well, the first method is that you could get your AI agent to set up a local model, install it, and then to have that as an option inside your AI agents. So you can see, for example, here, this is method number one.

Method number one is that you use the locally installed model, and then you basically say to your AI agents whether that's inside Codex or inside Claude, I need you to use that as a local model, so that inside Codex, we can use you as a cloud model, but we can also have the option of using local free models like LFM 2.5, 26B instead. And this doesn't have to be LFM, I'm just using that as an example because we've already installed it. So you can see an example right here.

And then what we can do here is we can say, can you create a web page using LFM 2.5, 26B? And then from there, you can basically command with prompts the cloud model to control the local model, which is pretty cool.

Now, there's another method as well. So what you can actually do is you could have, for example, Claude code or a local AI agent engine set up inside your agent OS, and then Claude or Codex could use that directly. So for example, I have free Claude code, and I have the local engine inside our agentic operating system. And Codex, the cloud model, or for example, Claude code could operate that at any time for us. The benefit of that is that basically it can control our OS for us, but we still get to use the cloud model. Now, you might say, okay, but I want to use, for example, Claude code with its own models as a drop down inside here. So you switch between the models and that sort of thing. The only method that I've seen that's really good for that is inside deep seek agent harness. And the reason that you would use deep seek harness is because everything inside is customizable. Every single element is fully open source as an agent harness. And what that means, for example, is you can quickly change the models and you can switch between, for example, like deep seek and then, for example, local models. So you can go inside creator mode, inside deep seek harness, and you can say, okay, allow me to use both local models as a drop down inside the model selector, as well as, for example, Astra 6, which is currently inside Codex, or, for example, Fable 5.1, which is currently inside Claude Code. And then you can actually create that as an option inside deep seek harness. Now, if you want to see that in action, you can see that inside our deep seek harness training over here. This is really the only harness where you can do that, right? And it gets a little bit more complex, but if you want more customization, obviously, you're adding more complexity as well. So that is method number three. And then the final method is just kind of like an interesting way as well. The other option that you have is you don't have to use local models. You could use a free API with your AI agents that they control. So for example, you could actually get GPT-6 Astra to control a free API for you.

As an example of that, we actually set up Token Harbor, which has access to deep seek V41 Flash for free for like two weeks, right? So you get a free API key with deep seek V41 Flash and Token Harbor. Now, why is that important? Because we basically have API key that we can control via Claude or via Codex. And then if we need to delegate work to that API, we can do it directly inside Codex.