wtf is claudex?

Claudex is a small setup script for developers who already use both the Claude Code command line tool and an OpenAI Codex subscription. It lets you run the Claude Code interface, with all its familiar commands and menus, while actually sending requests to OpenAI's models instead of Anthropic's, using the subscription you already pay for through Codex. The README explains a specific technical problem this project solves. If you are already logged into Claude Code with a Pro or Max subscription, a normal attempt to redirect it to a different model provider quietly fails, because Claude Code keeps using its own stored login token no matter what other settings you provide. The author found this out by inspecting the actual network traffic. The fix here is to give this alternate setup its own separate configuration folder, one that has no stored Claude login in it, so Claude Code is forced to use the alternate connection instead. Setting it up involves cloning the project and running its install script, which downloads a small local proxy program, sets up a background service that only listens on the local machine, and reuses your existing Codex login so you do not need to log in again separately. Once installed, running the command starts a normal-looking Claude Code session that is actually talking to the other provider's models behind the scenes, including a way to switch between different response depths or effort levels. The author is upfront that this routes a paid subscription through an unofficial local proxy, which sits in a legal gray area of that provider's terms of service, so it is used at the reader's own judgment. It currently only works on Linux, not macOS. It is released under the MIT license.