Run Codex on your own development machine, from anywhere.

Coding agents work best where the code actually runs

A coding agent is only as useful as the environment it can see. The version of your project that matters is the one on your development PC: the right Node or Python version, the local database, the .env file, the services already running on their ports.

The problem is that this machine is usually on your desk. When you are away from it, you either wait, reach for remote desktop, or try to recreate the environment somewhere else.

How it works

1. 1

   Connect the machine once

   Install the Dev Server Deck Local Agent on the Windows or Linux machine that holds your projects and register it with an Agent Token. macOS is coming soon.

2. 2

   Set up Codex from Studio

   Studio detects whether the Codex CLI is installed on that machine and can install it for you. Sign in with your ChatGPT account in the browser or with a device code - which also works from a phone.

3. 3

   Give it the task

   Open Studio on a registered project and describe the change. Choose the Codex model and reasoning level, attach screenshots or documents, and ask for a plan first when you want to review the approach.

4. 4

   Approve and follow along

   Codex's file changes and commands wait for your approval. You see its narration, reasoning and every step live, with diffs you can open in the editor.

You stay in control of every change

Codex runs against your working copy, so approvals matter. Studio keeps them explicit and fast.

• Approve or decline each command and file change as it is requested.
• Approve everything for the current turn, or for the session, when you trust the task.
• Open any changed file as a real red/green diff before you accept the result.

Long work keeps going when you close the tab

A turn runs on your machine through the Local Agent, not in your browser. Close Studio or lose your connection and the turn keeps working; reopen it to see the steps, narration and any pending approvals exactly where they are.

Goal mode for larger objectives

Set an objective and a token budget and let Codex keep working toward it turn after turn. It stops at the budget, a turn cap, or when it is done, and pauses for you on anything risky.

Not only Codex

Studio also runs Claude, GPT, Gemini, OpenRouter and other providers with your own keys, or local models through Ollama and LM Studio when code context should never leave the machine.

What you need to know

• The development machine must be powered on, awake and connected to the internet.
• A ChatGPT account with Codex access. Codex usage counts against that account.
• Studio traffic between your browser and the machine is end-to-end encrypted, so the Dev Server Deck relay cannot read it. Code context Codex needs is sent to OpenAI, as with any Codex session.

Questions

Does my project get uploaded to Dev Server Deck?

No. The Local Agent works on the project folder already on your machine. Studio messages travel end-to-end encrypted through the relay, which cannot read them. The only party that sees code context is the AI provider you choose - OpenAI, for Codex.

Can I use Codex from my phone?

Yes. Studio works in a mobile browser: you can start a task, read the live progress, and approve or decline commands from your phone while the work runs on your PC.

What happens if my browser disconnects mid-task?

Nothing is lost. The turn continues on the machine, and when you reopen Studio it resumes showing the steps, narration and any approval that is waiting for you.