Codex CLI Configuration Guide

Last updated: October 17, 2026 (Asia/Shanghai)

Related: Help Center · LLM API · VS Code

Applies to: using OpenAI Codex CLI with requests pointed at the tokenexplore gateway.

Key settings

Item	Value
Node.js	≥ 18 (20/22 LTS recommended)
Base URL	https://ai.afp.net/v1 (must include /v1, unlike Claude Code)
Protocol	Usually wire_api = "responses"; if incompatible, switch to the Chat Completions-related setting (depending on the Codex version)

INFO

Codex targets the OpenAI Responses / Chat Completions ecosystem. The tokenexplore model list is mainly Claude, Gemini, Kimi, DeepSeek, GLM, Qwen, Seedance, and so on — first use GET /v1/models to confirm which models your token can access, then set model in config.toml. If a model does not support Responses, run a minimal test before integrating, or switch to Claude Code / OpenCode.

---

1. Environment and token

node --version

Create a dedicated token in the console (e.g. named codex) and copy the sk- key.

---

2. Install

npm install -g @openai/codex
codex --version

---

3. Configuration files

~/.codex/config.toml

model_provider = "ciyuan"
model = "claude-sonnet-4-6"
model_reasoning_effort = "high"
disable_response_storage = true

[model_providers.ciyuan]
name = "tokenexplore"
base_url = "https://ai.afp.net/v1"
wire_api = "responses"
requires_openai_auth = true

Replace model with a model ID that your token can actually use and that has been tested to support the current wire_api. If Responses returns an error, consult the current Codex documentation and try changing wire_api to the Chat Completions mode it supports (field names may differ slightly between versions).

~/.codex/auth.json

{
  "OPENAI_API_KEY": "sk-xxxxxxxx"
}

On Windows, the path is usually %USERPROFILE%\.codex\.

---

4. Verify

codex "write a hello world"
# or specify a model (example)
codex --model claude-sonnet-4-6 "implement quicksort with tests"

---

5. Troubleshooting

Symptom	Fix
404 / can't connect	Make sure base_url includes /v1; restart the terminal
Model not found	Check against /v1/models; verify the token group and the available-model whitelist
Responses 4xx	The model may not support Responses → switch models, change wire_api, or use another CLI
Node too old	Upgrade to 18+ and reinstall
Insufficient quota	Top up in the console or raise the token quota

---

Back to: Help Center