gitwtfhub

wtf is codex-router?

duolahypercho/codex-router — explained in plain English

Analysis updated 2026-05-18

181JavaScriptAudience · developerComplexity · 4/5Setup · moderate

TL;DR

A local router that lets AI desktop apps like Codex and Claude Desktop use outside models such as Kimi, DeepSeek, and Grok through one credential-isolated setup.

Mindmap

mindmap
  root((codex-router))
    What it does
      Routes outside AI models
      Works with Codex and Claude
      Isolates credentials per app
    Tech stack
      Node.js router
      Python or uv
      Git for install
    Use cases
      Add Kimi DeepSeek Grok models
      Experimental Claude gateway
      Monitor usage on macOS tray
    Audience
      Developers
      AI power users
    Setup
      Run guided install script
      Choose and authenticate providers
      Restart the target app

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

Why would anyone build with this?

REASON 1

Add Kimi, DeepSeek, Grok, or Claude models to the Codex app's built in model picker.

REASON 2

Try Claude Desktop's experimental third-party inference gateway with alternate model providers.

REASON 3

Keep separate provider logins and API keys isolated per app instead of sharing one config.

REASON 4

Monitor usage and health for the active AI provider from a macOS menu bar panel.

What's in the stack?

JavaScriptNode.jsPython

How it stacks up

duolahypercho/codex-routerjianshuo/ccglassgaearon/react-transform-catch-errors
Stars181181183
LanguageJavaScriptJavaScriptJavaScript
Last pushed2016-04-18
MaintenanceDormant
Setup difficultymoderateeasyeasy
Complexity4/52/52/5
Audiencedeveloperdeveloperdeveloper

Figures from each repo's GitHub metadata at analysis time.

How do you spin it up?

Difficulty · moderate Time to first run · 30min

Requires Node.js 22.19+, Git, and either uv or Python 3.10+ with venv, each model provider needs its own login or API key.

The README does not state a license.

Wtf does this do

Codex Router lets you use AI models from other companies, such as Anthropic, Kimi, DeepSeek, and xAI, inside AI desktop apps that normally only work with their own built in models. It does this by running a small local program on your own computer that sits between the app and these outside models, keeping your login credentials for each provider separate and isolated from one another. It currently supports the Codex app and command line tool as a stable target, with experimental support for Claude Desktop and the Cursor editor. Installing support for one app does not change the settings of another, and the project is an independent community effort not affiliated with any of the companies whose models it connects to. Setup is meant to be guided rather than manual. You can either paste a prepared instruction into an AI coding agent so it installs everything for you, or run a provided install script for macOS, Linux, or Windows. The installer walks you through choosing which providers to connect, detects logins you already have, lets you sign in to some providers directly, and asks for API keys through a hidden terminal prompt rather than displaying them on screen. It does not make any billed test request unless you specifically ask it to. Using it requires the target app itself, a recent version of Node.js, either the uv tool or Python with a virtual environment, and Git. Once connected, supported models such as different Kimi, DeepSeek, Grok, and Claude versions appear in the app's own model picker, each tied to its own authentication method, which may be an existing account login or a separately billed API key. Codex Router adds only a small, clearly marked block to the app's configuration file so it can find the router, while leaving your existing accounts, profiles, and settings untouched. On macOS there is also an optional menu bar control panel that shows usage and health information for whichever provider you are actively using. Because Codex Router runs entirely on your own machine and stores keys locally with restricted file permissions, it is aimed at people comfortable running command line tools and installers, rather than a fully hosted service.

Yoink these prompts

Prompt 1
Walk me through installing the Codex target of codex-router using the guided install script.
Prompt 2
Explain how codex-router keeps my Kimi and DeepSeek API keys separate and secure on my machine.
Prompt 3
Help me enable a new provider and set its API key using the model-router command line tool.
Prompt 4
Show me how to troubleshoot models not appearing in Codex after installing codex-router, using the doctor command.

Frequently asked questions

wtf is codex-router?

A local router that lets AI desktop apps like Codex and Claude Desktop use outside models such as Kimi, DeepSeek, and Grok through one credential-isolated setup.

What language is codex-router written in?

Mainly JavaScript. The stack also includes JavaScript, Node.js, Python.

What license does codex-router use?

The README does not state a license.

How hard is codex-router to set up?

Setup difficulty is rated moderate, with roughly 30min to a first successful run.

Who is codex-router for?

Mainly developer.

View the repo → Decode another repo

This repo across BitVibe Labs

Don't trust strangers blindly. Verify against the repo.