Codex CLI - Yelu AI API

Codex CLI supports custom model providers. Configure a user-level yelu provider to send Responses API requests to https://api.yelu.ai/v1 with your Yelu API key.

The configuration below uses gpt-5.4, a coding-capable model recognized by the Yelu gateway. Model access is account-specific; confirm availability with GET /v1/models.

Configuration at a glance

Setting	Value
Provider ID	`yelu`
Base URL	`https://api.yelu.ai/v1`
Key environment variable	`YELU_API_KEY`
Wire API	`responses`
Example model	`gpt-5.4`
User configuration	`$CODEX_HOME/config.toml` or `~/.codex/config.toml`

Unlike Claude Code, Codex requires the /v1 suffix because its provider base URL is the OpenAI-compatible API root.

macOS

1. Install Codex

Use the official standalone installer:

macOS

curl -fsSL https://chatgpt.com/codex/install.sh | sh
codex --version

Homebrew and npm are also supported:

brew install --cask codex

npm install -g @openai/codex

2. Set the key

macOS

read -s "YELU_API_KEY?Yelu API key: "
echo
export YELU_API_KEY

The key remains available to processes started from this terminal and is not written to shell history. For persistent use, load it from macOS Keychain or another secret manager in your shell startup.

3. Create the provider configuration

Create ~/.codex/config.toml and add:

~/.codex/config.toml

model = "gpt-5.4"
model_provider = "yelu"
model_reasoning_effort = "medium"

[model_providers.yelu]
name = "Yelu AI API"
base_url = "https://api.yelu.ai/v1"
env_key = "YELU_API_KEY"
wire_api = "responses"

Provider settings must be in the user-level file. Codex intentionally ignores model_provider and model_providers in a project-local .codex/config.toml.

Windows

Native PowerShell

Install Codex using the official Windows installer:

Windows PowerShell

powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex"
codex --version

Set the key for the current session without echoing it:

Windows PowerShell

$secret = Read-Host "Yelu API key" -AsSecureString
$env:YELU_API_KEY = [System.Net.NetworkCredential]::new("", $secret).Password

Create %USERPROFILE%\.codex\config.toml:

Windows PowerShell

New-Item -ItemType Directory -Force "$HOME\.codex" | Out-Null
@'
model = "gpt-5.4"
model_provider = "yelu"
model_reasoning_effort = "medium"

[model_providers.yelu]
name = "Yelu AI API"
base_url = "https://api.yelu.ai/v1"
env_key = "YELU_API_KEY"
wire_api = "responses"
'@ | Set-Content -Encoding utf8 "$HOME\.codex\config.toml"

To make the key available in future terminals, use a Windows secret manager. If you choose a user environment variable for convenience, remember that it is stored as plain user configuration rather than in a dedicated secret vault. Native Codex on Windows uses the Windows sandbox. The default is appropriate for most projects; do not switch to unrestricted execution just to work around a project permission error.

WSL 2

Use WSL 2 when your repository and tools live in Linux:

Administrator PowerShell

wsl --install
wsl

Then install and configure Codex inside WSL:

WSL 2

curl -fsSL https://chatgpt.com/codex/install.sh | sh
read -rsp "Yelu API key: " YELU_API_KEY; echo
export YELU_API_KEY
mkdir -p ~/.codex
codex --version

Put the same TOML provider block in the WSL file ~/.codex/config.toml. The Windows and WSL Codex installations use separate home directories, keys, and configuration files. Keep WSL projects under ~/code instead of /mnt/c for better filesystem performance.

Verify the API and configuration

First confirm that the key can see the model:

curl -sS https://api.yelu.ai/v1/models \
  -H "Authorization: Bearer $YELU_API_KEY"

Invoke-RestMethod `
  -Uri "https://api.yelu.ai/v1/models" `
  -Headers @{ Authorization = "Bearer $env:YELU_API_KEY" }

Validate the local configuration and run a read-only smoke test from a Git repository:

Terminal

codex --strict-config --version
codex exec --strict-config --ephemeral --sandbox read-only \
  "Reply with exactly: Yelu Codex is ready"

Inside an interactive session, run /status to confirm that the provider is yelu, the model is gpt-5.4, and the active directory is correct.

Safe defaults

Keep YELU_API_KEY in an environment variable or secret manager; never place the key directly in TOML.
Start unfamiliar repositories with --sandbox read-only or use /permissions before allowing edits.
Use workspace-write only when the task should modify the current workspace.
Review the diff and run tests before accepting generated changes.
Avoid --dangerously-bypass-approvals-and-sandbox on a normal workstation.

Troubleshooting

Symptom	Fix
`YELU_API_KEY` is missing	Set it in the same terminal that launches Codex, then restart Codex.
Requests go to OpenAI	Check that `model_provider = "yelu"` is in the user-level `config.toml`, not only a project config.
`404` or malformed URL	Use `base_url = "https://api.yelu.ai/v1"` exactly.
Unknown model	Query `GET /v1/models` and set an enabled Responses-compatible coding model.
Config key rejected	Update Codex, then run with `--strict-config` to identify the unsupported key.
WSL cannot see the Windows key	Export `YELU_API_KEY` inside WSL; Windows and WSL environments are separate.
Native Windows sandbox fails	Run `codex doctor`; use WSL 2 if the native sandbox is incompatible with the workstation policy.

Next step

Continue with the Codex CLI examples for exploration, implementation, review, image input, and automation.

​Configuration at a glance

​macOS

​1. Install Codex

​2. Set the key

​3. Create the provider configuration

​Windows

​Native PowerShell

​WSL 2

​Verify the API and configuration

​Safe defaults

​Troubleshooting

​Next step

​Official references

Configuration at a glance

macOS

1. Install Codex

2. Set the key

3. Create the provider configuration

Windows

Native PowerShell

WSL 2

Verify the API and configuration

Safe defaults

Troubleshooting

Next step

Official references