OpenAI Codex CLI

Overview

Codex CLI is OpenAI’s official command-line programming assistant designed for terminal workflows. Integrating it with API易 takes one thing only:

Replace OpenAI’s endpoint with API易

API易 is an OpenAI-compatible interface (transparent proxy)—at its core, you just swap the Base URL and Key. No custom model_provider, no custom env_key required. If it works with the OpenAI SDK, it works with Codex CLI.

🚀 One-Step Setup

Two environment variables — no complex config files

⚡ Latest Models

Supports gpt-5.5 / gpt-5.4 / gpt-5.3-codex

💰 Same Billing Model

Aligned with OpenAI’s billing, better pricing

🔧 Cross-Platform

Mac / Linux / Windows (PowerShell) — all supported

1. Prerequisites

Install Codex CLI

Globally install the official CLI (requires Node.js 18+):

npm install -g @openai/codex

Verify:

codex --version

Mac users hitting permission errors can prefix with sudo; using nvm / fnm to manage Node is recommended to avoid global-install permission issues.

2. Core Config: Two Environment Variables

API易 is an OpenAI-compatible interface, so the entire config is two lines:

OPENAI_BASE_URL=https://api.apiyi.com/v1
OPENAI_API_KEY=sk-your-APIYI-key

Variable names are fixed: must be OPENAI_API_KEY and OPENAI_BASE_URL. These names are hardcoded in Codex CLI — do not rename them.Base URL must include /v1: write it as https://api.apiyi.com/v1. Missing /v1 will 404.

Mac / Linux Setup

Edit your shell config:

vim ~/.zshrc
# or ~/.bashrc

Add:

export OPENAI_BASE_URL="https://api.apiyi.com/v1"
export OPENAI_API_KEY="sk-your-APIYI-key"

Apply:

source ~/.zshrc
# or source ~/.bashrc

Windows Setup

✅ PowerShell (persistent, recommended)
Temporary (current session only)
GUI System Environment Variables

Use setx to write user-level environment variables:

setx OPENAI_BASE_URL "https://api.apiyi.com/v1"
setx OPENAI_API_KEY "sk-your-APIYI-key"

Close and reopen the terminal for it to take effect.

Effective only in the current PowerShell window:

$env:OPENAI_BASE_URL="https://api.apiyi.com/v1"
$env:OPENAI_API_KEY="sk-your-APIYI-key"

Good for quick tests.

Open “System Environment Variables”
Add user variables:
- Name OPENAI_BASE_URL, value https://api.apiyi.com/v1
- Name OPENAI_API_KEY, value sk-your-APIYI-key
Restart the terminal

3. Launch & Basic Usage

Enter your project and launch:

cd /your/project
codex

Or run a one-shot task:

codex "write a Python HTTP server for me"

Non-interactive (silent) mode:

codex -q "fix the build errors in this project"

4. Models (API易 Recommendations)

Common models available on API易 for Codex CLI:

Model	Strengths	Best For
`gpt-5.5`	Latest flagship	Complex coding, engineering analysis, agent workflows
`gpt-5.4`	Stable workhorse	Day-to-day coding, debugging, refactoring
`gpt-5.4-mini`	Cheap 5.4 variant	Mid-scale tasks, batch processing
`gpt-5.4-nano`	Ultra-cheap	Simple completions, lightweight tasks
`gpt-5.3-codex`	Codex-tuned	Complex software engineering tasks
`gpt-4.1`	Classic & cheap	Budget-sensitive daily use

How to pick: daily → gpt-5.4; heavy work / agents → gpt-5.5; pure code engineering → gpt-5.3-codex; cost-saving → gpt-5.4-mini / gpt-4.1.

1. Specify model at launch

Use -m or --model:

codex -m gpt-5.5
codex --model gpt-5.4
codex -m gpt-5.3-codex

You can also pass a task directly:

codex -m gpt-5.5 "review this project structure and point out areas to optimize"

2. Specify model in non-interactive mode

codex -q -m gpt-5.4 "fix the build errors in this project"

3. Switch models inside a session

After entering Codex CLI, in the active session:

/model

Then choose or type the model to switch to.

4. Configure default model

To make a model the default, edit ~/.codex/config.toml:

model = "gpt-5.5"

Codex’s current official config file is config.toml (not the older config.json). The user-level config lives at ~/.codex/config.toml; project-level .codex/config.toml is also supported.

5. Advanced Configuration

Custom System Prompts

Edit:

~/.codex/instructions.md

You can define coding style, output language, project conventions, e.g.:

- Write code comments in English
- Follow the project's ESLint config
- Provide detailed explanations

Project-Level AGENTS.md

On first entry into a project, run codex /init to generate AGENTS.md recording the project structure and conventions. To make Codex respond in a specific language by default, add:

Always respond to the user in English for this project.

Common Flags

codex -h          # full help
codex -m <model>  # specify model
codex -q          # non-interactive / silent mode
codex --full-auto # auto-execute (use cautiously)

6. Notes

1. Key & Variable Names

Get your API易 Key from the console at api.apiyi.com/token
Variable name must be OPENAI_API_KEY — hardcoded in the CLI, do not rename

2. Base URL

Must include /v1: https://api.apiyi.com/v1
Other gateway domains (e.g. b.apiyi.com/v1, vip.apiyi.com/v1) work the same way

3. Network Issues

If you hit timeouts / connection failures, check in order:

Local HTTP/HTTPS proxy settings
DNS resolution
Whether you’re accidentally using a Cloudflare relay domain (not recommended)

4. Model Differences

Codex CLI is biased toward coding tasks; multimodal capabilities (image/audio) are better called via dedicated endpoints
Different gpt-5.x models vary in tool calling, long-context handling, and reasoning depth — pick by task complexity

7. FAQ

Why does the OpenAI Codex CLI work with API易?

Because API易 is fully compatible with the OpenAI API protocol — https://api.apiyi.com/v1 and https://api.openai.com/v1 are interchangeable in request/response format. Replacing the Base URL is enough.

`command not found: codex`

Confirm install:

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

If it still fails, check whether npm bin -g is on your PATH.

Invalid API Key (401 / Invalid Key)

Make sure you’re using an API易 Key (starts with sk-), not an OpenAI key

Verify the env var is actually exported:

echo $OPENAI_API_KEY        # Mac/Linux
echo %OPENAI_API_KEY%       # Windows CMD
echo $env:OPENAI_API_KEY    # Windows PowerShell

After changing env vars, restart the terminal or run source ~/.zshrc

Connection error / timeout / 404

Most common cause: Base URL is missing /v1.Correct: https://api.apiyi.com/v1Next, check your local proxy and DNS.

How do I switch / upgrade models?

Three ways:

One-off: codex -m gpt-5.5
In session: /model
Default: edit ~/.codex/config.toml, set model = "gpt-5.5"

Which models are supported?

OpenAI series: ✅ fully supported (recommended gpt-5.5 / gpt-5.4 / gpt-5.3-codex)
Other vendor aggregated models: supported by API易, but Codex CLI itself is biased toward the OpenAI protocol — non-OpenAI models may behave differently around tool use / function calling
For Claude / Gemini-based coding, use their native CLIs (e.g. Claude Code)

Suitable for production?

CLI: best for dev-time productivity
Production: prefer direct API calls (more control, monitoring, gradual rollout)

How do I uninstall or disable the API易 setup?

Uninstall the CLI:

npm uninstall -g @openai/codex

Disable API易 config: remove the env vars and config file.

# Mac/Linux: edit ~/.zshrc or ~/.bashrc and remove the OPENAI_* lines
rm ~/.codex/config.toml

# Windows: remove OPENAI_BASE_URL / OPENAI_API_KEY from System Environment Variables

8. Summary

The whole integration is one sentence:

Replace OpenAI’s endpoint with API易

The two essential lines:

export OPENAI_BASE_URL="https://api.apiyi.com/v1"
export OPENAI_API_KEY="sk-your-APIYI-key"

Everything else — model picking, prompts, instructions.md, AGENTS.md — is just polish on top.

API易 Console

Manage API keys and view usage

Claude Code Integration

Use Claude models for CLI coding

Model Comparison

All available models and pricing

API Manual

General calling conventions

Overview

AI Agent

Conversational AI

Programming

Engineering

Translation

Community Contributions

Overview

🚀 One-Step Setup

⚡ Latest Models

💰 Same Billing Model

🔧 Cross-Platform

1. Prerequisites

Install Codex CLI

2. Core Config: Two Environment Variables

Mac / Linux Setup

Windows Setup

3. Launch & Basic Usage

4. Models (API易 Recommendations)

1. Specify model at launch

2. Specify model in non-interactive mode

3. Switch models inside a session

4. Configure default model

5. Advanced Configuration

Custom System Prompts

Project-Level AGENTS.md

Common Flags

6. Notes

1. Key & Variable Names

2. Base URL

3. Network Issues

4. Model Differences

7. FAQ

8. Summary

API易 Console

Claude Code Integration

Model Comparison

API Manual

Overview

AI Agent

Conversational AI

Programming

Engineering

Translation

Community Contributions

Documentation Index

​Overview

🚀 One-Step Setup

⚡ Latest Models

💰 Same Billing Model

🔧 Cross-Platform

​1. Prerequisites

​Install Codex CLI

​2. Core Config: Two Environment Variables

​Mac / Linux Setup

​Windows Setup

​3. Launch & Basic Usage

​4. Models (API易 Recommendations)

​1. Specify model at launch

​2. Specify model in non-interactive mode

​3. Switch models inside a session

​4. Configure default model

​5. Advanced Configuration

​Custom System Prompts

​Project-Level AGENTS.md

​Common Flags

​6. Notes

​1. Key & Variable Names

​2. Base URL

​3. Network Issues

​4. Model Differences

​7. FAQ

​8. Summary

​Related Resources

API易 Console

Claude Code Integration

Model Comparison

API Manual

Overview

1. Prerequisites

Install Codex CLI

2. Core Config: Two Environment Variables

Mac / Linux Setup

Windows Setup

3. Launch & Basic Usage

4. Models (API易 Recommendations)

1. Specify model at launch

2. Specify model in non-interactive mode

3. Switch models inside a session

4. Configure default model

5. Advanced Configuration

Custom System Prompts

Project-Level AGENTS.md

Common Flags

6. Notes

1. Key & Variable Names

2. Base URL

3. Network Issues

4. Model Differences

7. FAQ

8. Summary

Related Resources