How to configure OpenAI Codex CLI to use Azure AI Foundry (Azure OpenAI)

codex-with-azure.md

Configuring OpenAI Codex CLI with Azure AI Foundry

This guide explains how to configure the OpenAI Codex CLI to use Azure AI Foundry (Azure OpenAI) as the model provider.

Prerequisites

• OpenAI Codex CLI installed (npm install -g @openai/codex or via other methods)
• An Azure AI Foundry resource with a deployed model
• Your Azure OpenAI API key

Setup

1. Set your API key as an environment variable

export AZURE_OPENAI_API_KEY="your-api-key-here"

Add this to your shell profile (~/.zshrc, ~/.bashrc, etc.) for persistence.

2. Create or edit ~/.codex/config.toml

# Set your model deployment name and provider
model = "gpt-5.1-codex-max"  # Replace with your actual Azure deployment name
model_provider = "azure"

# Optional: Set reasoning effort for reasoning models
model_reasoning_effort = "high"

# Configure the Azure provider
[model_providers.azure]
name = "Azure OpenAI"
base_url = "https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1"
env_key = "AZURE_OPENAI_API_KEY"
wire_api = "responses"

Key Configuration Notes

Field	Description
model	Your Azure deployment name (not the model name). This is the name you gave when deploying the model in Azure.
model_provider	Must be "azure" to use the Azure provider defined below.
base_url	Your Azure OpenAI endpoint. Important: Do NOT include /responses at the end - Codex appends it automatically.
env_key	The environment variable name containing your API key.
wire_api	Use "responses" for the new Responses API (required for newer models like GPT-5.1, o-series, etc.).

3. Trust your project directory (optional)

To avoid being prompted each time, add trusted directories:

[projects."/path/to/your/project"]
trust_level = "trusted"

Finding Your Azure Endpoint

1. Go to Azure AI Foundry or Azure Portal
2. Navigate to your Azure OpenAI resource
3. Find the endpoint URL (looks like https://YOUR-RESOURCE-NAME.openai.azure.com/)
4. Append /openai/v1 to get the base_url

Example Complete Config

model = "gpt-5.1-codex-max"
model_provider = "azure"
model_reasoning_effort = "high"

[model_providers.azure]
name = "Azure OpenAI"
base_url = "https://my-azure-resource.openai.azure.com/openai/v1"
env_key = "AZURE_OPENAI_API_KEY"
wire_api = "responses"

[projects."/Users/myuser/projects/myapp"]
trust_level = "trusted"

Testing

Run a simple test:

codex exec "say hello"

If you encounter issues, enable debug logging:

RUST_LOG=debug codex exec "test"

Troubleshooting

"OperationNotSupported" Error

• Your model may not support the chat completions API. Use wire_api = "responses" for newer models.

Double /responses/responses in URL

• Remove /responses from your base_url. Codex appends it automatically when wire_api = "responses".

Authentication Errors

• Verify AZURE_OPENAI_API_KEY is set correctly
• Check that your API key has access to the deployment

References

• Azure OpenAI Responses API Documentation
• OpenAI Codex CLI