DEV Community

Cover image for Crazyrouter Codex CLI: Use Codex with One API Key and an OpenAI-Compatible Gateway
Jenny Met
Jenny Met

Posted on • Originally published at crazyrouter.com

Crazyrouter Codex CLI: Use Codex with One API Key and an OpenAI-Compatible Gateway

#ai #webdev #tutorial #tools

Crazyrouter Codex CLI: Use Codex with One API Key and an OpenAI-Compatible Gateway

OpenAI Codex CLI is useful when you want an AI coding agent directly inside your terminal. The painful part is not the idea — it is the setup: API keys, base URLs, model names, Windows environment variables, macOS shell profiles, Linux config files, and different providers for different models.

The new crazyrouter-codex-cli repo solves one specific problem:

connect Codex CLI to Crazyrouter with one API key and an OpenAI-compatible API endpoint.

Repository:

https://github.com/xujfcn/crazyrouter-codex-cli
Enter fullscreen mode Exit fullscreen mode

What this repo does

The repo provides simple install scripts for:

  • Windows PowerShell
  • Windows batch file
  • macOS
  • Linux

It configures Codex CLI to use:

OPENAI_BASE_URL=https://cn.crazyrouter.com/v1
OPENAI_API_KEY=your-crazyrouter-key
Enter fullscreen mode Exit fullscreen mode

That means Codex CLI can talk to Crazyrouter through an OpenAI-compatible interface, while Crazyrouter handles the model/provider side.

Important rule:

Do not add UTM parameters to API endpoints. UTM belongs on human-clickable website links, not OPENAI_BASE_URL.

Correct:

export OPENAI_BASE_URL=https://cn.crazyrouter.com/v1
Enter fullscreen mode Exit fullscreen mode

Wrong:

export OPENAI_BASE_URL=https://cn.crazyrouter.com/v1?utm_source=...
Enter fullscreen mode Exit fullscreen mode

Why use Codex CLI through Crazyrouter?

A terminal coding agent is most useful when it can become part of your normal development loop:

  1. open a project directory;
  2. ask the agent to inspect code;
  3. let it patch files;
  4. run tests;
  5. review the diff;
  6. repeat.

But real developer teams often want more than one model. Some tasks need a fast low-cost model. Some need a stronger reasoning model. Some need Claude-style code review. Some need Gemini-style long-context analysis. Some teams also need a more stable route from regions where direct access is unreliable.

Crazyrouter gives Codex CLI a single OpenAI-compatible gateway:

  • one API key;
  • one base URL;
  • multiple model choices;
  • OpenAI-compatible client configuration;
  • easier switching between coding models.

One-command install

Windows PowerShell

Open PowerShell as a normal user and run:

iwr -UseB https://raw.githubusercontent.com/xujfcn/crazyrouter-codex-cli/main/install-crazyrouter-codex.ps1 | iex
Enter fullscreen mode Exit fullscreen mode

Or download and run:

install-crazyrouter-codex.bat
Enter fullscreen mode Exit fullscreen mode

macOS / Linux 

curl -fsSL https://raw.githubusercontent.com/xujfcn/crazyrouter-codex-cli/main/install-crazyrouter-codex.sh | bash
Enter fullscreen mode Exit fullscreen mode

The script asks for your Crazyrouter API key, writes the needed environment variables, and backs up existing Codex configuration when needed.

Manual setup

If you prefer to configure it yourself, install Codex CLI first:

npm install -g @openai/codex
Enter fullscreen mode Exit fullscreen mode

Node.js 22+ is recommended.

Then set the environment variables.

macOS / Linux 

export OPENAI_API_KEY=sk-your-crazyrouter-key
export OPENAI_BASE_URL=https://cn.crazyrouter.com/v1
Enter fullscreen mode Exit fullscreen mode

Windows PowerShell 

setx OPENAI_API_KEY "sk-your-crazyrouter-key"
setx OPENAI_BASE_URL "https://cn.crazyrouter.com/v1"
Enter fullscreen mode Exit fullscreen mode

After setx, reopen your terminal.

Then start Codex:

codex
Enter fullscreen mode Exit fullscreen mode

Codex config.toml example

Some Codex CLI versions support provider configuration in:

  • Windows: %USERPROFILE%\.codex\config.toml
  • macOS / Linux: ~/.codex/config.toml

Example:

model = "gpt-5.5"
model_provider = "crazyrouter"

[model_providers.crazyrouter]
name = "Crazyrouter"
base_url = "https://cn.crazyrouter.com/v1"
env_key = "OPENAI_API_KEY"
wire_api = "responses"

[model_providers.crazyrouter.query_params]
Enter fullscreen mode Exit fullscreen mode

If you see this error:

wire_api = "chat" is no longer supported
Enter fullscreen mode Exit fullscreen mode

change:

wire_api = "chat"
Enter fullscreen mode Exit fullscreen mode

to:

wire_api = "responses"
Enter fullscreen mode Exit fullscreen mode

Model selection

Once the gateway is configured, you can start Codex with the default model or specify one explicitly:

codex
codex --model gpt-5.5
codex --model gpt-4o-mini
codex --model claude-sonnet-4-6
Enter fullscreen mode Exit fullscreen mode

Model availability can vary by account, provider route, and current upstream status. Check the current model list here:

Crazyrouter model list

Practical workflow example

A simple Codex CLI coding loop:

cd your-project
codex
Enter fullscreen mode Exit fullscreen mode

Then ask:

Inspect this repo and explain the architecture. Do not edit files yet.
Enter fullscreen mode Exit fullscreen mode

After the summary:

Find the smallest safe fix for the failing login test. Make the change, then run the relevant test only.
Enter fullscreen mode Exit fullscreen mode

Then review:

git diff
npm test -- login
Enter fullscreen mode Exit fullscreen mode

The gateway setup does not replace careful review. It makes the model/provider configuration less annoying so you can focus on the actual engineering loop.

Troubleshooting

codex: command not found

Install Codex globally:

npm install -g @openai/codex
Enter fullscreen mode Exit fullscreen mode

Then check:

which codex
codex --help
Enter fullscreen mode Exit fullscreen mode

On Windows, reopen the terminal after installation.

API key not found

Check whether the environment variable exists.

macOS / Linux:

echo $OPENAI_API_KEY
Enter fullscreen mode Exit fullscreen mode

Windows PowerShell:

$env:OPENAI_API_KEY
Enter fullscreen mode Exit fullscreen mode

If empty, set it again.

Wrong base URL

The base URL should be exactly:

https://cn.crazyrouter.com/v1
Enter fullscreen mode Exit fullscreen mode

Do not add /chat/completions, /responses, or UTM parameters. Client libraries append the final API path themselves.

Existing Codex config conflict

If you had another provider configured before, check:

cat ~/.codex/config.toml
Enter fullscreen mode Exit fullscreen mode

Make sure the selected model_provider points to the Crazyrouter provider block.

When this setup is useful

This repo is especially useful for developers who:

  • use Codex CLI but want an OpenAI-compatible gateway;
  • want to try multiple models from one CLI setup;
  • work across Windows, macOS, and Linux machines;
  • need a repeatable install script for teammates;
  • want a simpler path for AI coding workflows in regions where direct provider access can be unstable.

Links

Bottom line

crazyrouter-codex-cli is a small repo, but it removes a common setup tax: configuring Codex CLI to use an OpenAI-compatible gateway correctly.

If you want Codex CLI with one key, one base URL, and easier model routing, start here:

https://github.com/xujfcn/crazyrouter-codex-cli
Enter fullscreen mode Exit fullscreen mode

Top comments (0)