jujishou/codex-switch

codex-switch is a small Windows desktop app that lets you redirect the OpenAI Codex App to use other AI models instead of OpenAI's own. By default, Codex only works with official OpenAI models. This tool intercepts those requests locally and forwards them to whichever provider you choose, such as DeepSeek, Kimi, Zhipu, or Tongyi. The practical effect is that you can keep using the Codex interface you already know while paying much lower API rates, or using Chinese-hosted services that do not require a proxy connection. The app works by running a small local proxy on your machine that translates the Codex request format into the standard OpenAI Chat Completions format that most alternative providers support. When you switch to a non-OpenAI model, the proxy starts automatically and Codex routes its traffic through it. When you want to go back to the original OpenAI setup, one click stops the proxy and restores the original configuration. The switch is clean: it merges only the specific fields it needs to change in the Codex config file, leaving your projects, plugins, and marketplace settings untouched. Several providers are bundled out of the box, including two DeepSeek variants and four Kimi variants. If you want to add another service, the app includes nine preset templates covering common platforms such as Zhipu, Tongyi, Doubao, Groq, Mistral, and OpenRouter. You pick a template, the base URL fills in automatically, and you only need to paste your API key. Fully custom entries are also supported if your provider is not on any template. Each provider's key is stored separately, so switching between them does not overwrite anything. The app ships as a single .exe file that requires no Python installation. It has a modern light and dark UI that follows your system theme. On first run it backs up your existing Codex config file before making any changes, so you have a fallback. The proxy only listens on localhost and is not reachable from outside your machine. At the moment only one upstream provider can be active at a time, which is a limitation of how Codex itself works, not of this tool.