OpenAI Codex CLI

Configure the OpenAI Codex CLI to use KiosAPI as the backend provider with the Codex model groups.

OpenAI Codex CLI

OpenAI Codex CLI is a command-line coding agent from OpenAI. This guide walks you through configuring it to work with KiosAPI.

System Requirements

Requirement	Minimum	Recommended
Operating System	macOS 12+, Ubuntu 20.04+, or Windows 11 (WSL2)	—
RAM	4 GB	8 GB
Node.js	22+	Latest LTS
Disk Space	200 MB	500 MB

Get the right API key

Codex models on KiosAPI (gpt-5.5, gpt-5.6, gpt-6 series) are served through the Codex-Pro and Codex-Plus groups:

1. Log in to the KiosAPI Dashboard.
2. Go to Console → Tokens → Add Token.
3. Select group Codex-Pro (0.04×) or Codex-Plus (0.02×) — check which group covers the exact model you want on the pricing page.
4. Save and copy the key (starts with sk-).

A token in the wrong group cannot call the Codex models. Codex-Pro and Codex-Plus cover slightly different model sets — verify your target model’s group on the pricing page before creating the key. See Choosing a Token Group.

Installation

Install Codex CLI globally via npm:

npm i -g @openai/codex

Or via Homebrew (macOS/Linux):

brew install codex

Verify the installation:

codex --version

Current Codex models on KiosAPI

Model	Group	Best for
gpt-6-astra	Codex-Pro, Codex-Plus	Most capable agentic coding
gpt-5.6-sol	Codex-Pro, Codex-Plus	Daily coding — recommended default
gpt-5.6-terra	Codex-Pro, Codex-Plus	Balanced coding
gpt-5.5	Codex-Pro	Lighter tasks

Configuration

1. Create the config file

Create ~/.codex/config.toml:

[model_providers.kiosapi]
name = "KiosAPI"
base_url = "https://kiosapi.com/v1"
wire_api = "responses"
env_key = "KIOSAPI_KEY"

[profiles.kiosapi]
model = "gpt-5.6-sol"
model_provider = "kiosapi"

The wire_api = "responses" setting tells Codex to use OpenAI’s Responses API format, which KiosAPI supports at /v1/responses.

2. Set the environment variable

Set your KiosAPI key as the KIOSAPI_KEY environment variable.

Windows (PowerShell):

setx KIOSAPI_KEY "sk-xxx"

macOS / Linux:

Add to ~/.bashrc or ~/.zshrc:

export KIOSAPI_KEY="sk-xxx"

Then reload:

source ~/.bashrc  # or ~/.zshrc

Usage

Interactive mode

codex

Direct question

codex "Explain how async/await works in Python"

Using the KiosAPI profile

codex --profile kiosapi "Write a function to reverse a linked list"

VSCode Plugin

Codex is also available as a VSCode extension. After installing the extension, configure it to use KiosAPI by setting the same environment variables in your VSCode settings or terminal.

The VSCode extension reads environment variables from your shell. Make sure KIOSAPI_KEY is exported in the shell that launches VSCode, or set it in the integrated terminal.

Advanced Configuration

Non-Interactive CI Mode

For CI/CD pipelines, run Codex in non-interactive mode:

codex --quiet --auto-approve "Fix the failing test in tests/auth.test.js"

Flag	Description
--quiet	Suppress interactive prompts
--auto-approve	Automatically approve file changes
--json	Output results as JSON

MCP Servers

Codex supports Model Context Protocol (MCP) servers for extended tool access. Add MCP servers in ~/.codex/config.toml:

[mcp.filesystem]
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/dir"]

[mcp.github]
command = "npx"
args = ["-y", "@modelcontextprotocol/server-github"]
env = { GITHUB_PERSONAL_ACCESS_TOKEN = "ghp_xxx" }

Troubleshooting

Authentication Error (401)

• Verify KIOSAPI_KEY is set correctly: echo $KIOSAPI_KEY
• Ensure the key starts with sk-
• Check the env_key in your config matches the environment variable name

Model Not Found / No Available Channel

• Verify the model name in your profile is a current model ID (old names like o4-mini no longer exist)
• Verify your token’s group (Codex-Pro or Codex-Plus) covers that model on the pricing page

Connection Error

• Ensure base_url is https://kiosapi.com/v1 (with /v1)
• Check your network connection and firewall settings

Verify Your Setup

codex "Say hello"

Need Help?

• KiosAPI Dashboard
• Choosing a Token Group
• Available Models