How to Use OpenAI Codex CLI your own AI API Key?

The Codex CLI is a powerful interface for interacting with AI in your terminal. However, locking yourself into a single provider limits your potential.

By connecting the Codex CLI to Zeabur AI Hub, you can route your queries to any top-tier model (like Claude, Grok, or Gemini) using the exact same tool. This setup works on a pay-as-you-go basis—no need to juggle multiple subscriptions for different AI providers.

Here is how to set it up.

🧱 Prerequisites

1. Install Codex CLI:

   npm install -g @openai/codex
   
   

2. Zeabur Credentials:

   • Go to your Zeabur Dashboard to get your AI Hub API Key.

3. Choose your Endpoint

   Zeabur currently provides two endpoints. Pick the one geographically closest to you:

   • Asia (Tokyo): https://hnd1.aihub.zeabur.ai/
   • US West (San Francisco): https://sfo1.aihub.zeabur.ai/

---

⚙️ Quick Setup Guide

You need to configure Codex to talk to Zeabur's unified API instead of a specific provider directly.

1. Create the Configuration File

Create or edit the config file at ~/.codex/config.toml. Paste the following configuration to define Zeabur as a provider:

# Default model to use
model = "gemini-3-pro-preview"
model_provider = "zeabur"

[model_providers.zeabur]
name = "Zeabur AI Hub"
# Note: Ensure you append /v1 to the endpoint
base_url = "<https://hnd1.aihub.zeabur.ai/v1>"
env_key = "ZEABUR_API_KEY"
wire_api = "chat"

2. Set your API Key

You need to make your API key available to the tool.

Option A: Just trying it out (Temporary)

For a single session usage:

export ZEABUR_API_KEY="your-zeabur-api-key"
codex "Who are you?"

Option B: Set it and forget it (Permanent)

To avoid setting the key every time, add it to your shell profile.

1. Open your shell configuration (e.g., ~/.zshrc or ~/.bashrc):

   nano ~/.zshrc
   
   

2. Paste this line at the bottom:

   export ZEABUR_API_KEY="your-zeabur-api-key"
   
   

3. Save and reload: source ~/.zshrc

---

💸 Start Coding (With Any Model)

Now you can use the CLI tool with the flexibility to switch models on the fly.

Basic Usage:

codex "Write a hello world function in Python"

Switching Models:

You can override the default model defined in your config using the -m flag.

# Use Gemini
codex -m gemini-3-pro-preview "hi"

# Use Claude
codex -m claude-sonnet-4-20250514 "hi"

# Use GPT-4o
codex -m gpt-4o "hi"

Why this is better:

• Flexibility: Switch between Grok, Claude, and GPT without changing tools.
• Cost Efficiency: Pay only for the tokens you use via Zeabur, rather than paying fixed monthly fees for multiple AI subscriptions.

---

🔧 Troubleshooting

If you run into issues:

• 401 Unauthorized:
  • Check that base_url in your config includes /v1.
  • Verify your environment variable is set: echo $ZEABUR_API_KEY.
• Model Not Found:
  • Ensure the model name (e.g., gpt-4o) is currently supported by Zeabur AI Hub. Check the Zeabur dashboard for the list of available model IDs.

Happy Coding!