Coding tools

Codex CLI

Connect OpenAI's local coding agent Codex CLI to RuToken through a custom model provider.

Codex CLI is OpenAI's coding agent that runs locally in your terminal (GitHub). It edits files, runs commands under sandbox and approval policies and uses the OpenAI Responses API, which RuToken supports at https://rutoken.ai/v1/responses.

Prefer a guided setup? The /codex-setup page generates the configuration and commands for your system in one click.

Prerequisites

An API key: open the console, go to API Keys and create a dedicated key (for example, named Codex) with access to the models you plan to use.
Node.js 22 or newer (for the npm installation), or Homebrew on macOS.
On Windows, native support is available; OpenAI also recommends WSL2 for the best experience.

Installation

npm

npm install -g @openai/codex

Homebrew (macOS)

brew install --cask codex

Check the installation:

bash

codex --version

Configuration

1. Edit config.toml

Open the Codex configuration file (create the folder and the file if they do not exist):

OS	Path
macOS / Linux	`~/.codex/config.toml`
Windows	`%USERPROFILE%\.codex\config.toml`

Merge the following. model and model_provider must be at the top level, before any [section]:

toml

model = "gpt-5-codex"          # replace with a model ID from the Model Square
model_provider = "newapi"

[model_providers.newapi]
name = " RuToken"
base_url = "https://rutoken.ai/v1"
env_key = "NEWAPI_API_KEY"
wire_api = "responses"

Key	Value
`base_url`	`https://rutoken.ai/v1` — Codex appends `/responses`
`env_key`	name of the environment variable that holds your key
`wire_api`	keep `"responses"`

2. Set the API key

macOS / Linux

export NEWAPI_API_KEY="sk-xxxx"
# To keep it, add the same line to ~/.zshrc or ~/.bashrc

Windows (PowerShell)

$env:NEWAPI_API_KEY = "sk-xxxx"
# To keep it for new terminals:
setx NEWAPI_API_KEY "sk-xxxx"

$env: only lasts for the current window, and setx only affects windows opened afterwards. Start codex from a terminal where the variable is set.

Verify

bash

cd /path/to/your/project
codex

Choose the approval mode when asked, then send Only reply "Connection successful". Do not read or write files or run commands. You can also test non-interactively:

bash

codex exec "Reply with OK"

Use /model inside Codex to switch models. Each request appears in the console under Usage Logs.

Troubleshooting

Symptom	Fix
`NEWAPI_API_KEY` is missing	Launch `codex` from the terminal where the variable is set; `env_key` must match the variable name exactly.
`401` / invalid API key	Copy the full key again; check that it is enabled, not expired and has quota left, then set the variable again.
`404` / Responses not supported	`base_url` must end with `/v1`, and `wire_api` must be `"responses"`. Make sure the chosen model supports the Responses API.
Model not found	Replace `model` with an ID from the Model Square and check the key's model limits.
Another provider is still used	Put `model` and `model_provider` at the top of the file; check whether a `profile` or project-level config overrides them.

← Previous

Claude Code

Factory Droid CLI