📌 About Codex CLI Codex CLI is OpenAI's official terminal coding agent, written in Rust, capable of reading, modifying, and running code locally (open source). It connects to OpenAI by default, but supports custom model_provider, so it can seamlessly connect to the LLMoxy gateway and use GPT-5.4 and other models.

Installation Steps

1. Install Node.js

Visit the Node.js official website. We recommend downloading the LTS version on the left, which is the long-term support version and more stable.

2. Install Codex CLI

Open a command line and install globally via npm:

npm install -g @openai/codex

macOS users can also use Homebrew:

brew install codex

Verify installation:

codex --version

Codex CLI supports macOS, Linux, and Windows. Windows users are recommended to run natively in PowerShell, or use WSL2 for a native Linux environment.

Configuration Steps

Codex CLI's configuration file is located at ~/.codex/config.toml (Windows: %USERPROFILE%\.codex\config.toml). Connecting to LLMoxy only requires two steps: register a custom provider + provide API Key.

1. Get API Key

Go to the LLMoxy Console → API Key Management → Copy secret key.

Note: Please use a token from the auto group.

2. Write config.toml

Add the following to ~/.codex/config.toml (create if it does not exist):

# Default to LLMoxy provider and model
model = "gpt-5.4"
model_provider = "llmoxy"

[model_providers.llmoxy]
name = "LLMoxy"
base_url = "https://llmoxy.com/v1"
wire_api = "responses"
env_key = "LLMOXY_API_KEY"

Configuration item explanations:

model — Default model name, fill in a model supported by LLMoxy (e.g., gpt-5.4, gpt-5.3-codex)
model_provider — Points to the provider ID defined below
base_url — LLMoxy gateway address, with /v1
wire_api — Interface protocol. LLMoxy supports both /v1/responses (Responses API, Codex native mode, recommended) and /v1/chat/completions (Chat Completions). Prefer responses; if compatibility issues occur, change to chat
env_key — Environment variable name for storing the API Key, set in the next step

TIP Unlike Claude Code, Codex CLI's base_url must include /v1.

3. Set API Key environment variable

The environment variable specified by env_key needs to be set with a real Key in the system.

Windows (PowerShell, permanent):

[Environment]::SetEnvironmentVariable("LLMOXY_API_KEY","Your LLMoxy API Key","User")

After setting, close PowerShell and open a new window for it to load.

macOS / Linux:

{
echo ''
echo '# LLMoxy'
echo 'export LLMOXY_API_KEY="Your LLMoxy API Key"'
} >> ~/.zshrc   # Bash users change to ~/.bashrc
source ~/.zshrc

Verify:

echo "$LLMOXY_API_KEY"

You must replace Your LLMoxy API Key with the real Key copied from the LLMoxy Console.

4. Launch

Enter your project directory and start an interactive session:

codex

If you can chat and execute tasks normally, you have successfully connected to LLMoxy.

Common Operations

Switch model and reasoning intensity

Enter /model in the session to switch models or adjust reasoning effort. You can also specify at launch via -m:

codex -m gpt-5.3-codex

Approval modes

Codex requests confirmation before editing files or executing commands. You can switch in the session with /approvals, or specify at launch:

codex --ask-for-approval      # Confirm each step (default, safest)
codex --full-auto             # Auto-execute, only within working directory in a network-restricted sandbox

💡 Tip When running for the first time in a project directory, Codex will ask whether you trust that directory.

Scripting (non-interactive)

Use the exec subcommand to hand one-off tasks to Codex, suitable for calling in scripts or CI:

codex exec "Add unit tests for ParseConfig in main.go"

Image input

You can attach screenshots or design drafts to let Codex understand requirements combined with images:

codex -i screenshot.png "Implement the login page according to this design draft"

Web search

Enable Web search to get the latest information:

codex --search

Connect MCP tools

Register MCP services in config.toml via [mcp_servers.*] to provide third-party tools and context for Codex. This configuration can coexist with the LLMoxy provider configuration:

[mcp_servers.chrome-devtools]
command = "npx"
args = ["chrome-devtools-mcp@latest"]

Common Issues

`npm install -g` permission error

Do not use sudo npm install -g. We recommend using nvm to manage Node.js, or modify the npm global directory:

npm config set prefix ~/.npm-global

Then add ~/.npm-global/bin to PATH. macOS users can also install Node.js via Homebrew (no sudo needed).

Still asks to log in to OpenAI account after launch

Confirm model_provider = "llmoxy" is active in config.toml, and the LLMOXY_API_KEY environment variable is set. After configuring a custom provider, Codex uses API Key authentication and does not require ChatGPT login.

Error 404 / model does not exist

Confirm base_url includes /v1: https://llmoxy.com/v1
Confirm model is filled with a model name supported by LLMoxy
If wire_api = "responses" errors, try changing to chat

Environment variables not taking effect

Windows: SetEnvironmentVariable(...,"User") requires reopening the terminal
macOS / Linux: run source ~/.zshrc (or ~/.bashrc), or reopen the terminal, then verify with echo "$LLMOXY_API_KEY"

API call failure / balance issue

Check that the API Key has no extra spaces
Go to the LLMoxy Console to check account balance and token group (recommended to use auto group)

Remote server connection timeout (Linux)

Check network: curl -I https://llmoxy.com
If proxy is needed, configure the HTTPS_PROXY environment variable

Codex CLI Configuration Guide