​

OpenAI Codex With Your Own Model

​

Codex In Action With OpenResponses

​

Key Benefits

• Run Codex with Custom Models: Integrate OpenAI Codex with OpenResponses to use any model of your choice
• Extend Functionality: Enhance Codex capabilities with additional MCP tools or custom tools and integrations
• Simple Deployment: Quick setup with no separate installation needed - just follow the quickstart guide
• Full Control: Maintain complete ownership of your code data and model choices

​

Step-by-Step Setup Instructions

​

1. Run OpenResponses Service

Launch OpenResponses using Docker:

docker run -p 8080:8080 masaicai/open-responses:latest

For advanced configuration options, refer to the detailed quickstart guide.

​

2. Install OpenAI Codex

Install the Codex CLI globally:

npm i -g @openai/codex

For more information about Codex CLI, visit the official repository.

​

3. Configure and Run Codex with Your Preferred Model

Set OpenResponses as your base API and configure your API key:

# Point to your OpenResponses instance
export OPENAI_BASE_URL=http://localhost:8080/v1

# Configure your API key (use the key from your model provider, e.g., OpenAI, Claude, etc.)
export OPENAI_API_KEY=your-api-key-here

Run Codex with a specific model using the -m flag:

​

Example with Locally Deployed Model

codex -m "http://mymodel_host/v1@my_model" "explain me the structure of the codebase"

​

Example with Claude

codex -m "claude@claude-3-5-haiku-20241022" "explain me the structure of the codebase"

​

Example with DeepSeek

codex -m "deepseek@deepseek-chat" "analyze this repository"

​

Example with Google Gemini

codex -m "google@gemini-2.0-flash" "help me understand this code"

The format for specifying a model is provider@model-name or model_endpoint@model_name, where:

• provider: The model provider (e.g., claude, deepseek, google, openai)
• model_endpoint: For locally deployed models or any custom model provider, the endpoint URL where chat/completions is available
• model-name: The specific model to use from that provider or endpoint

​

Supported Model Providers

OpenResponses supports the following model providers that can be used with the provider@model_name convention:

For any model provider not listed in the table above, you can use the model_endpoint@model_name convention by directly specifying the full API endpoint URL. This works for both locally deployed models and any third-party API services that follow the OpenAI-compatible chat/completions format.