OpenAI-Compatible Models

This local Codex plugin adds the openai_compatible_chat tool. It sends one chat-completions request to the provider configured on this computer. It does not change Codex's own model.

Configure a provider

Create %APPDATA%\OpenAICompatibleModel\config.json with a provider URL and model ID. Example for Ollama:

{
  "base_url": "http://localhost:11434/v1",
  "model": "qwen2.5:14b",
  "api_key_env": "OPENAI_COMPATIBLE_API_KEY"
}

For a hosted provider, use its documented OpenAI-compatible API base URL and model ID. Set its API key in the Windows user environment variable named by api_key_env. For example, set OPENAI_COMPATIBLE_API_KEY in Windows Environment Variables, then restart Codex. Leave that environment variable unset for local providers that do not require a key.

The plugin reads configuration on each request. It accepts HTTP or HTTPS base URLs, including localhost for local model servers. It appends /v1/chat/completions unless the URL already ends in /v1 or /chat/completions. Redirects are rejected. Do not put API keys in chat or in this plugin's source files.

Install

The openai-compatible-models marketplace folder contains the local plugin source. Install the marketplace from that folder with Codex's plugin marketplace command, then install openai-compatible from that marketplace. Restart Codex after installation. The marketplace source folder must remain in place while the plugin is installed.