Quick note before I get into it: I'm Jimmy, I build small Mac tools, and this one is the most embarrassingly simple thing I've shipped in a while. It is also the one I use the most. Fair warning.
I have a confession that I think a lot of people who switched from Codex to Claude Code have been quietly sitting on. The model is better. The agentic behavior is better. The tool calls feel sturdier. But every single time I open a terminal and start typing, some small part of my brain misses the way Codex used to lay things out. The collapsible diffs. The way file edits felt like a real surface, not a stream of text scrolling by. The little keyboard rhythms. I tried to talk myself out of it for weeks. I just couldn't.
So I built Claudx. It is Claude Code underneath, with a Codex style UI sitting on top. You point it at a repo and it behaves like the agent you already use, but the screen finally feels familiar again.
Why bother
I almost didn't ship this. I assumed someone at one of the bigger players would do it first, or that I would get over the muscle memory in a week. Neither happened. Around day ten of typing in raw Claude Code I noticed I was rereading the same diff three times because the formatting wasn't what my eyes were trained on. That is a real productivity tax, even if it sounds silly written out.
Tools should bend to how you already work, not the other way around. I have been using coding agents seriously since the first Codex preview, and at this point my hands know certain shapes. Asking me to relearn the shapes every time the underlying model gets better is the wrong tradeoff. The model is the hard part. The UI is the easy part. So I made the easy part match the muscle memory and let the hard part stay where it is.
What it actually does
Claudx wraps Claude Code and presents the session through a panel layout that should feel instantly recognizable to anyone who spent real hours in Codex. Diffs are collapsible. File edits get their own surface instead of disappearing into a chat scroll. Long tool outputs are folded by default so you can scan, not drown. The keybindings are the ones your fingers already know.
Underneath, nothing exotic is happening. It is Claude Code. Same agent loop, same tool calls, same model selection. I am not trying to fork the agent, reinvent the protocol, or wedge myself between you and Anthropic. I am just rendering the session in a way that does not fight my brain.
A few things I deliberately kept simple. There is no account system. No telemetry pipeline. No cloud sync. You install it, you unlock it once, and it runs locally on top of your existing Claude Code setup. If Claude Code keeps shipping (and it will), Claudx rides along. I am not trying to build a moat, I am trying to scratch an itch.
What I learned shipping it
A few honest observations from the build, in case you are sitting on a similar itch.
First, the gap between "this is annoying me" and "this is worth building" is smaller than you think. I kept waiting for the feeling to fade. It did not. After about two weeks of complaining to myself I realized the complaint was the spec.
Second, wrapping an existing tool is wildly underrated as a starting point. I did not have to retrain anything. I did not have to evaluate models. I did not have to think about prompts. The hard work was already done by the team at Anthropic. My job was just to render the output in a format that respects how I actually read code.
Third, shipping small is freeing. I have a bad habit of wanting every project to be a platform. Claudx is not a platform. It is one binary that does one thing. The smaller I made the surface area, the faster I shipped, and the less I second-guessed myself.
Fourth, the people who care about tiny UI details are not a niche. Every engineer I have shown this to has the exact same reaction within about forty seconds, which is some flavor of "oh thank god." The market for the same tool but laid out the way I am used to is much bigger than the market for a brand new tool you have to learn.
Who this is for
If you switched from Codex to Claude Code and your hands still reach for the wrong keys, this is for you. If you never used Codex and you are happy with raw Claude Code, you do not need this and I am not going to try to convince you otherwise. If you are on the fence, the demo video on claudx.org walks through a real session end to end.
I am one person building this in my spare time alongside a few other small Mac tools. Bug reports and feature requests come straight to me. That is the whole pitch. I missed a UI, I rebuilt it, I am using it every day, and now you can too.
Top comments (0)