Codex CLI Setup for AMD LLM Gateway

This repository contains an automated setup script for installing and configuring Codex CLI to work with AMD's LLM Gateway.

Overview

Codex CLI is OpenAI's command-line tool for AI-assisted coding. This setup script configures it to use AMD's LLM Gateway, supporting multiple model profiles including o3, GPT-5, Claude Sonnet 4, and Gemini 2.5 Pro.

Features

• 🚀 Automated Installation: Interactive setup with confirmation prompts
• 🔧 Multi-Profile Configuration: Support for o3, GPT-5, Claude, and Gemini models
• 📁 Portable Installation: Installs locally in current directory
• 🛠 Shell Integration: Automatic shell alias setup for bash, csh, and tcsh
• ✅ Dependency Checking: Validates Node.js and npm availability

Prerequisites

• Node.js 18+: Required for Codex CLI
  • Install from https://nodejs.org/en/download
• npm: Node package manager (usually included with Node.js)
• AMD LLM API Key: Valid API key for AMD's LLM Gateway
• Linux/Unix Environment: Script designed for bash shell environments

Quick Start

1. Clone or download this repository

2. Run the setup script:

   ./setup-codex-cli.sh

3. Set your API key in your shell config file (automatically added):

   • For bash: Edit ~/.bashrc
   • For csh: Edit ~/.cshrc
   • For tcsh: Edit ~/.tcshrc

   Replace your-api-key-here with your actual API key.

4. Start using Codex:

   codex

Installation Options

The setup script provides:

• Interactive Mode (default): Prompts for confirmation before installation
• Auto-confirm Mode: Use --yes or -y flag to skip prompts

Both modes install Codex CLI in the current directory and create shell aliases for easy access.

Configuration

Model Profiles

The setup creates multiple model profiles in ~/.codex/config.toml:

• o3 (default): Uses AMD Gateway with responses API
• gpt5: GPT-5 via AMD Gateway with responses API
• claude: Claude Sonnet 4 via Babel local gateway
• gemini: Gemini 2.5 Pro via Babel local gateway

Switch profiles using:

codex --profile gpt5
codex --profile claude
codex --profile gemini

Environment Variables

Set your AMD LLM API key:

For bash users (add to ~/.bashrc):

export AMD_LLM_API_KEY='your-api-key-here'

For csh/tcsh users (add to ~/.cshrc or ~/.tcshrc):

setenv AMD_LLM_API_KEY 'your-api-key-here'

Gateway Endpoints

• AMD Gateway: https://llm-api.amd.com/openai/{model}
• Babel Local Gateway: http://localhost:5000/v1 (optional, for Claude/Gemini)

Usage Examples

# Start interactive Codex session with default (o3) model
codex

# Use a specific profile
codex --profile gpt5
codex --profile claude

# Get help
codex --help

# Check version
codex --version

File Structure

codex/
├── setup-codex-cli.sh    # Main setup script
├── .npmrc                # npm configuration (created during setup)
├── package.json          # NPM dependencies (created during setup)
└── README.md             # This file

What the Setup Script Does

1. Directory Confirmation: Verifies installation location
2. Environment Setup: Loads module system (if available)
3. Node.js Validation: Checks for Node.js 18+ and npm
4. npm Configuration: Sets up local npm directories
5. Package Installation: Installs @openai/codex via npm
6. Configuration: Creates ~/.codex/config.toml with multi-profile setup
7. Shell Integration: Adds aliases and API key placeholders to shell configs

Troubleshooting

Node.js Version Issues

If you get Node.js version errors, install Node.js 18+ from nodejs.org or use a version manager like nvm.

API Key Issues

Verify your AMD_LLM_API_KEY is set correctly:

echo $AMD_LLM_API_KEY

Babel Gateway for Local Models

To use Claude or Gemini profiles, ensure Babel gateway is running on localhost:5000.

Module System

The script attempts to use the Pandora module system if available. If not found, it falls back to checking system PATH for Node.js.

Model-Specific Notes

o3 (Default)

• Uses AMD Gateway responses API
• Full tool/function calling support

GPT-5

• No temperature parameter support (uses default 1.0)
• Use max_tokens instead of max_completion_tokens
• Full tool/function calling support

Claude Sonnet 4

• Requires Babel local gateway
• Uses chat API format

Gemini 2.5 Pro

• Requires Babel local gateway
• Uses chat API format