Codex is now FREE: Here’s how…

**Julian Goldie** (0:00)
Today, I want to show you how to use Codex for free. And this is a super powerful system you can basically use to code across like 93 different coding tools with one single tool, and then you can plug it into Claude Codex. If you're not sure what that means, let me just show you. So this is Codex. And what we've done is we've routed it through OmniRoute, which means we've got 93 providers plugged into Codex, and we don't need to subscribe to Codex to use it. So for example here, if we say, okay, build out a beautiful landing page for an SEO agency, we can actually code for free using this powerful system. And we can also see the things that we've coded out previously. So this is our Agent OS system, we've got Codex inside there, and then we've used OmniRoot inside Codex to build and automate whatever we want. So you can see that's working right now. And then if we switch over to this tab, we can actually see the things that we've previously created. So you can see, for example, here we actually created a demo, as you can see here, and then we've got the full HTML page here for another test we did.
And so we can build and code whatever we want, and it actually looks pretty good. So it's using the harness of Codex, but it's free to use because we've got OmniRoot plugged into that system. So today I'm going to walk you through exactly how it works step by step, with what I call the free Codex engine. So it's OpenAI's Codex coding agent, but it's rooted through OmniRoot, across 93 providers with auto fallback when it runs dry. So like for example, if one free API doesn't work, no problem, you can switch to a different one. And so Codex is still creating for you, but the brain of it is with OmniRoot, which is pretty amazing stuff. So if you're not familiar with OmniRoot, it's just dropped recently. It's already got 12,000 stars on GitHub. I've already tested out inside OmniRoot directly. So for example, we've got OmniRoot as a coding tool built in. But if you want to use the full harness of Codex with OmniRoot, you can. And then also, if we have a look over here, you can see the finished landing page that we created here. The other cool thing I like about this is when you build with this stuff, I mean, it looks pretty sleek. It looks nice. The colors are nice. It's simple. It's not going to be like Fable 5 level, but at the same time, like the landing page is not too bad at all, considering that was a super basic prompt. That's because we're using the harness of Codex to get better outputs. The other thing I would say here is it's fairly quick. I mean, it only took about 60 to 90 seconds to create that, and everything I've built with it so far has looked really good.
You can also see the preview here as well. When you're building out inside the chat of Codex, we can see the preview, and this also links inside our Agent Operating System. You might also say, why would you use it with an Agent Operating System because then you can plug in the memory, the context, you can take what you've created with Codex, you can put it into Claude Code. You could, for example, design a landing page, but then plug in a video from the video agent inside the Agent OS. And so it all links together, it all syncs together. And also like, I don't need to open up Codex and then Claude separately. They all work inside one beautiful system. And we can do it for free. Like, that's pretty amazing to live in a world like that. So OmniRoot is super powerful, super fun to use, easy to use as well. 93 providers in there. It's never failed on me when I've tested it as well. And you can basically just code for free, unlimited. Now, the other thing I would say here is like, when you're using, for example, free models with Codex, you can run them locally. For example, like you can go over to Alarmr and you could get a free model like Laguna XS 2.1, and you could plug that locally into something like Codex. But from what I've seen, the outputs are much nicer from these APIs that it's using rather than just using the local models directly. It's also a lot faster. If you're using local models with something like Codex or Free Claude Code, the problem is that quite often, it uses up a lot of power on your computer. So it's much slower when it's responded. With this setup, it doesn't seem to slow anything down at all, and we just get really nice outputs as you can see here. So how does this work step-by-step? So essentially what we have is we have Codex, that goes through OmniRoute, it picks and falls over if one provider doesn't work, you get 93 providers and then you get everything written. So it's the same agent, there's no new tool to learn, it's a real Codex CLI, it plans, it writes files, it runs commands on its own exactly as before, it's only the backend that's changing. And also for me personally, when I've used GPT 5.5, one of the reasons I stopped using Codex itself was because it just ran out of tokens really quickly, like even on the page subscription. So that's why I think this might even be a better alternative in some ways. Then you can use a free pull. So instead of the OpenAI meter, Codex routes to 93 providers. 11 of them are actually permanent, free as well, which is pretty crazy.