LifeOS with chatgpt codex?

[HereForTheVibes]

Has anyone tried to use LifeOS with Codex or any other harnesses/models aside from Claude Code? What's your experience been like?

[danielmiessler]

Short answer: not drop-in. LifeOS is authored against Claude Code specifics: hooks, @-imports, the skills system, and the claude binary for inference, so Codex can't run it as-is. The transferable parts are the concepts and file structures (TELOS, ISAs, the docs), which any capable harness can read. Some folks bridge mixed-harness setups by message bus instead; see the discussion in #1493 for a working pattern that spans Claude Code and other agents. Making the installer harness-agnostic is a direction we care about, but no Codex port exists today.

[danielmiessler]

Yes, it's not an instant drop-in, but the whole system is designed to be used within your existing system and installed by your existing system. You can choose what pieces to take, what to install, and how to integrate it with your own live system, regardless of your underlying harness. The answer is yes, it's definitely possible and much easier than in previous versions, but it won't be as easy as if you were on Claude Code.

[icryptan]

https://claude.ai/code/artifact/e0a0e13e-6f60-4615-8e4b-3a8b8e3d1c41#no_universal_links

U can check this. I tried it, and it work, but it have problems. For example gpt send me message, that context overload after first message

[jdrolls]

Codex, no luck for me, but there are other solutions.

For GPT models, I use Pi a good deal because it's much more customizable than the Codex CLI tool. I've got it porting into the same LifeOs setup as my Claude. The main thing is that the plugins can work basically the same as hooks in Claude Code. You'd can use your codex sign in there. Sol 5.6 works great as a driver if you hit your anthropic limits.

Having your agent make a plan and port it over just takes maybe a couple hours. You may hit a few bugs initially, but they iron out pretty quick.

If you would like, I can have my agent share a few more details on what was done for the port. I would share the actual code, but I'm afraid at this point my version of LifeOS forks pretty far from the original framework, and many of my hooks aren't the same.

I tried the same thing with OpenCode a while back, and I know there was a forked project that did the same, but I have had much better luck with Pi for the harness to build on top of.

If you want a perhaps more elegant or easier solution there's an open source project called Claudex that allows you to use your OpenAI sign in Claude code. Then you aren't re-engineering a whole system. It actually works great, I just had to tweak around a bit to fix the context issue.

[HereForTheVibes]

Thanks! This is really useful. Have you found fable/opus to be better than sol 5.6? Or anything different with 5.6 sol? Openai are more generous with their limits so wanted to see if it's worth it or not

[jdrolls]

I've used it primarily when I run out of my Claude sub for the week. It works fine, I like the anthropic model feel better. I like using gpt for subagents and opus as primary driver personally, just like LifeOs is set up to do.

But, you should it a try for a month and see what you think! For 90% of what most people do, GPT-5.6 sol is going to be just fine if the harness is equal, at least that has been my experience. And yeah, it's significantly more generous in the sub.

Fable is amazing but is so much more expensive I use it sparingly. I like Opus a lot, and it fits my workflows well, but I honestly don't know that it's "better" than GPT except in certain specific tasks.

[danielmiessler]

You can technically install LifeOS into any harness because it is your receiving AI itself that's doing the install, and it's an expert on your system. So you can work with your AI to decide what components to bring over and what to leave alone.