Codex CLI

Codex CLI supports custom providers — describe a provider with our base_url in config.toml and point the key at an env var.

Install

You'll need Node.js 18 or newer:

terminal

npm install -g @openai/codex

Grab a cai-... key from API keys. Codex's config lives in ~/.codex/ (on Windows — %USERPROFILE%\.codex\).

Custom provider in config.toml

Create or extend ~/.codex/config.toml:

~/.codex/config.toml

model = "gpt-4o"
model_provider = "cheapai"

[model_providers.cheapai]
name = "cheapai.io"
base_url = "https://api.cheapai.io/v1"
env_key = "CHEAPAI_API_KEY"
wire_api = "chat"

The base_url already includes /v1 — that's the OpenAI-compatible endpoint. For model use any model from the catalog that supports chat.

Key via env var

The config.toml above sets env_key = "CHEAPAI_API_KEY" — Codex reads the key from that variable:

macOS / Linux · ~/.zshrc

export CHEAPAI_API_KEY="cai-..."

Windows · PowerShell

[Environment]::SetEnvironmentVariable("CHEAPAI_API_KEY", "cai-...", "User")

Alternative: ~/.codex/auth.json with the key in it; the env var is simpler and harder to commit by accident.

Run

terminal

codex            # interactive mode
codex "fix the failing test in src/auth"   # one-off task

Switch model for a single run: codex -m claude-3-5-sonnet (if the model is in the catalog and supports chat). Permanently — change model in config.toml.

Next: Cursor, Cline, Aider… · OpenAI-compatible API · Error codes.