How to Run OpenAI Codex Inside Google Antigravity
Photo by Benjamin Dada on Unsplash

If you’re a developer, you’re likely juggling a Gemini 3.1 Pro subscription for its 1M context window, Claude Opus 4.6 and a ChatGPT Plus account for its raw coding logic.

But why are you switching tabs?

While we already know our “perfect” IDE is giving us free Opus and Gemini 3.1, we have found a technical backdoor to use Codex too :)

By using a specific extension deep-link, you can run OpenAI Codex directly inside the Google Antigravity environment.

So, here’s how to use Google’s superior browser orchestration with OpenAI’s superior coding brain:

The Installation Secret (antigravity:extension/openai.chatgpt)

If you search the Antigravity marketplace for “ChatGPT” or “Codex,” you might get a “No results found” screen.

On support forums, people speculate this is Google’s way of keeping the ecosystem pure.

To bypass the gatekeepers, you need to use the deep-link protocol.

Copy and paste this directly into your browser:

antigravity:extension/openai.chatgpt

This triggers the installation of the OpenAI Codex IDE Extension.

Once you sign in with your OpenAI credentials, you’ll see the Codex logo in your sidebar.

You now have a “Dual-Brain” setup.

codex extension

“Skills” over Prompts: The Matrix Method

The biggest mistake devs make with Codex is treating it like a chat bot.

Instead of typing a 500-word prompt every time, you create a skills.md file in your project root.

  • The Theory: You install documentation or design rules into the AI’s short-term memory.
  • The Practice: I dropped the documentation for the Sarvam AI API into my skills.md. Now, when I want to mock a friend's persona, I just tag @skills in my Codex chat, and it knows exactly how to handle the local Hinglish nuances without me explaining it again.

The Workflow: Logic vs. Execution

Here is how I’m currently building my sports-tech webapp using this dual setup:

  • The UI (Gemini 3.1/Opus): Until I exhaust them.
  • The Logic (Codex): I use the OpenAI extension to architect the backend. Codex is currently considered the king of “clean code” and system design, it doesn’t generate the “AI slop” look that some earlier models were prone to.
  • The Execution (Antigravity Agent): Once Codex writes the code, I let the native Antigravity Agent handle the heavy lifting. It opens the terminal, runs my FastAPI server, and uses the integrated browser to test the UI.

The “Quota” Leverage

The real reason this is a “War Time” hack? Quota management.

As we discussed last week, Gemini 3.1 Pro High is currently eating 10% of your quota per prompt.

If you use Codex for the routine refactors and UI tweaks, you preserve your high-reasoning Google quota for the truly “impossible” research tasks and 1M token context audits.

left-panel
In 2026, loyalty to a single AI provider is a career bottleneck.

In case we are meeting for the first time, come over here, it’ll be worth the roller coaster of articles that are gonna come up in the next few weeks.