What is codex-desktop-custom-models?

Codex Desktop Custom Models is a starter kit for people who use Codex Desktop, a desktop AI coding assistant, and want to run a second, completely separate instance of it that connects to a different AI model provider. Instead of modifying your existing Codex setup, this tool creates a fully isolated clone with its own settings, data directory, icon, and launcher, running alongside the original. The reference implementation included in the kit sets up Codex MiniMax, a second Codex Desktop app with a red icon that connects to the MiniMax M2.7 model through a local bridge. That bridge is the key technical piece. Codex Desktop speaks one API format called the Responses API, but many third party model providers speak a different format called OpenAI-compatible Chat Completions. The bridge translates between these two shapes in real time, including handling tool calls. Setup is done through shell scripts that run on macOS and Linux. On macOS the installer creates a native app bundle and a launch agent, and on Linux it creates a desktop shell profile. Both platforms end up with a separate home directory, for example a folder named .codex-minimax, so the custom profile's sessions, config, and state never mix with the original Codex installation. The kit also includes a local automation server bundled with the custom profile, which supports scheduled background tasks and same thread follow up automations, since custom profiles do not automatically inherit host provided automation tools. The MiniMax profile is just a reference example. The same scripts and bridge can be adapted to any OpenAI-compatible provider by swapping the base URL, model name, and API key. The project is released under the MIT License.