Skip to main content
Codex is an AI coding tool that runs in your terminal. You can connect Codex to Novita’s OpenAI-compatible LLM API endpoint and use Novita models for code understanding, generation, refactoring, and debugging tasks.
Codex currently supports the Responses API only. Configure wire_api = "responses" as shown below, and do not switch this integration to Chat Completions mode.
Prefer one command? Novita Connect starts Codex on Novita models with novi codex, without editing config.toml.

​
Prerequisites

Before you begin, prepare the following values:
  • API key: Create or copy an API key from Key Management.
  • Base URL: https://api.novita.ai/openai/v1
  • Model ID: Copy the model ID you want to use from the Model Hub, such as deepseek/deepseek-v4.1-flash.
Keep your API key private. Use it only in your local Codex configuration and never commit it to a Git repository.

​
Install Codex

Before installing Codex, make sure Node.js 18 or higher is installed on your machine. On macOS, if you use Homebrew, you can install Node.js with:
Check your Node.js and npm versions:
Install the Codex CLI:
Check the Codex version:

​
Configure your API key

Open your shell configuration file, for example ~/.zshrc:
Add the following environment variable:
Reload your shell configuration and confirm the variable is available:

​
Configure Codex

Open the Codex configuration file:
Add the following configuration, and replace model with the Novita model ID you want to use:

​
Start Codex

Go to your project directory and start Codex:
You can now enter a coding task, for example:
If Codex returns an authentication or model error, check that:
  • NOVITA_API_KEY is available in the current terminal session.
  • model exactly matches the Model ID in Novita’s Model Hub.
  • base_url is set to https://api.novita.ai/openai/v1.
Last modified on October 15, 2026