Install Codex CLI and Connect It Through Lumecoder
Linux / Ubuntu / macOS / Windows: install the CLI, paste the console config, pay with points
After this setup, Codex CLI bills your Lumecoder account.
Connect it our way
These four steps are enough.
- Create an API key in the Lumecoder console and keep a balance
- Install Codex CLI
- Paste the two console snippets into ~/.codex
- Run codex in a project directory and send a message
Lumecoder is a third-party access service, not an official OpenAI product. Models and prices follow the console.
What you get
You still use Codex CLI on your machine. Login and billing go through your Lumecoder account. Copy the two snippets as-is. Do not edit them.
Get a key and a balance first
Without a key or with an empty wallet, the CLI starts but cannot complete a turn.
- Create a Lumecoder account
- Open the API console and mint an sk- key
- Read pricing, then top up before a real run
Install the CLI (short)
The binary is still OpenAI's Codex CLI. Use the first command on Linux, Ubuntu, and macOS; use the second on Windows. Open a new terminal and confirm which codex prints a path. If you use npm, the package is only @openai/codex.
curl -fsSL https://chatgpt.com/codex/install.sh | sh
powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex"
Homebrew cask and GitHub Release binaries exist too. OS requirements follow OpenAI's docs (currently macOS 12+, Ubuntu 20.04+ / Debian 10+, or Windows 11 via WSL2). On Ubuntu, do not feed npm an ancient apt Node.
Write the Lumecoder files
macOS / Linux (Ubuntu, WSL included) use ~/.codex/. Native Windows uses %USERPROFILE%\.codex\. Running codex /status once will create the directory.
Keep this at the top of config.toml:
model_provider = "lumecoder" model = "gpt-5.4" model_reasoning_effort = "medium" network_access = "enabled" disable_response_storage = true windows_wsl_setup_acknowledged = true model_verbosity = "medium" [model_providers.lumecoder] name = "lumecoder" base_url = "https://api.lumecoder.com" wire_api = "responses" requires_openai_auth = true
Replace sk-xxx in auth.json with the console key:
{
"OPENAI_API_KEY": "sk-xxx"
}Start and check
Start it from a repository. If the first-run wizard offers another sign-in, dismiss it and keep the files you already wrote.
cd /path/to/your/repo codex
One successful reply means the turn billed against Lumecoder. /status shows the model and directory. Use codex resume to return to the last chat.
Mistakes that break this setup
Using another sign-in from the wizard
That overwrites the files you just pasted. Keep the two console snippets and nothing else.
Pasting an OpenAI platform key
Use the key from the Lumecoder console. Do not swap in an OpenAI platform key or edit the console snippets.
Empty wallet
Correct files still fail with auth or balance errors. Check points in the console first.
Snippet appended at the bottom
Keep this block at the top of the file. A second login config below it will take over.
FAQ
Do I need Node.js on Linux or Ubuntu?
Not for the official script. Node 18+ is only for npm. The Lumecoder files are the same on every distro: ~/.codex.
Windows native vs WSL — which home directory?
Write the files in the environment that runs codex. WSL Ubuntu uses ~/.codex, not Windows %USERPROFILE%.
Can I change the default model?
Edit model in config.toml. The sample uses gpt-5.4 to match the docs page. Use an ID the console actually serves.
How do I get the last chat back?
Sessions stay local. Use codex resume or --last. Later turns still bill Lumecoder.