How to Connect GPT, DeepSeek, and Grok to Codex Using OpenCodex

The author explains a practical problem: needing GPT for complex refactoring, DeepSeek for low‑cost batch edits, and Grok for other tasks, and how to manage these models without repeatedly reconfiguring Codex.

What OpenCodex Does

OpenCodex acts as a local model gateway. It intercepts Codex requests and forwards them to the selected provider, allowing official and third‑party models to appear together in Codex’s model list.

Key benefits:

Different models can coexist, enabling task‑specific switching.

A visual configuration page shows providers, model names, endpoints, and connection status.

API keys and subscription logins are managed centrally.

Step 1 – Install

Node.js 18+ is required. Run the following command in a terminal: npminstall -g @bitkyc08/opencodex If the script is blocked by Bun, use:

npminstall -g --allow-scripts=bun @bitkyc08/opencodex

Verify the installation with: ocx --version After the first use, run initialization:

ocx init

Step 2 – Launch

Start the visual manager with: ocx gui The GUI opens the provider configuration page; if the service isn’t running, it will be started automatically.

Step 3 – Add DeepSeek

In the Providers page, click “Add provider” and search for DeepSeek. After selecting it, confirm the name and adapter, then fill in the API key, endpoint, and default model. The adapter used is openai-responses. Save and verify the connection status.

Subscription accounts can be added either directly on the Providers page or via command line. For example, to log in to an xAI account: ocx login xai After successful authorization, the model appears in Codex’s model list alongside the API key.

Step 4 – Use in Codex

Reopen Codex and open the model selection box. If GPT, DeepSeek, and Grok are all visible, the multi‑model configuration is active. Switch models just like with the built‑in provider.

The author recommends assigning tasks: use GPT for architecture design, complex refactoring, and tricky errors; use DeepSeek for batch edits, formatting, and simple scripts, e.g., switching between DeepSeek’s flash and vision models.

Model selection happens inside Codex, reducing cost and friction. If you already use Codex and have multiple model accounts, this configuration is worth trying.

Final Note

AI tools like Codex still have many pain points—phone verification, plus recharge, plugin usage, workflow building—each can stall. OpenAI’s policies change frequently, so the tutorial may need updates in the future.