Codex adds remote connections for Mac mini devboxes in the ChatGPT app
OpenAI documented Codex remote connections, letting the ChatGPT app point at a separate Codex host such as a Mac mini or rented VPS. Try it for long runs that need to stay alive off-device or for phone-first coding sessions.

TL;DR
- OpenAI's launch post put Codex into the ChatGPT mobile app as a preview on iOS and Android, while OpenAIDevs said the same rollout works with laptops, Mac minis, devboxes, and remote environments.
- OpenAI also published the launch post and remote connections docs, which turn a separate Codex host into a persistent machine you can steer from another desktop or from your phone, according to thsottiaux's note and Reibmachine's walkthrough.
- The first wave of usage clustered around always-on Macs: mattshumer_ repurposed a Mac mini as a Codex devbox, gdb called that setup "the way," and Reibmachine's Reddit post described the same pattern with a remote Mac mini and phone.
- The rollout is a little ahead of the polish. altryne's error screenshot showed failed authorization, thsottiaux acknowledged reset propagation issues on some accounts, and testingcatalog's reply said the adjacent Locked Use toggle was not broadly available.
You can read OpenAI's announcement, browse the remote connections docs, and compare that clean official framing with altryne's failed authorization screenshot, koltregaskes' Windows report, and testingcatalog's Locked Use reply. The interesting bit is not mobile alone. It is that Codex is starting to look like a long-lived host you keep running somewhere else, then poke from whatever screen you have handy.
Remote connections
OpenAI framed the feature as mobile continuity, but the more specific change lives in the remote connections docs: a Codex instance running on another machine can be exposed back into the ChatGPT app and controlled as a remote project.
That makes the host machine the durable part of the workflow. In Reibmachine's Reddit setup, Codex runs on a Mac mini with codex --yolo, then the same session is reached from a MacBook and an iPhone. OpenAI's own product copy in OpenAIDevs makes the same point from the other direction, saying Codex keeps running on the computer while the phone handles review, approvals, and new prompts.
A few details from the docs and evidence matter more than the marketing line:
- The host can be a laptop, Mac mini, devbox, or managed remote environment, per OpenAIDevs and the launch post.
- Remote SSH is already generally available, which is why the mobile feature immediately spills into rented boxes and office machines, according to OpenAIDevs.
- The ChatGPT app is not just a notification surface. OpenAI said in the launch post that you can review outputs, steer execution, approve commands, change models, and start new work from the phone.
Setup flow
The setup sequence is simple enough that the evidence pool repeats the same shape across posts. OpenAIDevs said to update both the ChatGPT mobile app and the Codex macOS app, then start setup from the Codex desktop sidebar.
Reibmachine's walkthrough fills in the missing steps with the actual terminal flow:
- Start Codex on the host machine with
codex --yolo. - Complete the device login in the browser and trust the working directory.
- Pair the phone through the ChatGPT app's Codex mobile flow.
- Authorize the desktop in the Connections screen.
- Add that machine as a remote project, so prompts from phone and laptop hit the same running session.
The UI screenshots add one more useful detail. testingcatalog's Connections screenshot shows a dedicated Connections settings page for authorizing a Mac to control other signed-in devices, while kevinkern's screenshot includes the warning that Codex will access the connected desktop's files, apps, and browser.
Always-on devboxes
The strongest signal in the reactions is that people immediately mapped the feature onto spare Macs and home servers. mattshumer_ said he wiped a Mac mini to turn it into an always-on devbox for Codex mobile, and mattshumer_'s follow-up said the payoff was closing the laptop while a run kept going.
That same host-first pattern shows up across the evidence:
- Reibmachine used a Mac mini as the permanent Codex box, with phone and laptop as controllers.
- altryne's update said a machine originally bought for OpenClaw or Hermes could now double as a remote Codex machine.
- npew's iPad post described managing sessions on both a Mac mini and a VPS.
- TheRealAdamG's repost of a multi-machine screenshot showed multiple hosts surfacing inside the mobile app.
- steipete's repost showed an app view listing several connected computers at once.
The phrase that kept recurring was some variation of "satellite device." OpenAIDevs' repost used it explicitly, and gdb boiled the pitch down to building from your phone while Codex lives somewhere else.
Windows and feature flags
The cleanest buried caveat is that several adjacent capabilities were on different rollout schedules. OpenAI's launch thread originally said phone-to-Windows support was coming soon, but by the next day koltregaskes said remote control did work on Windows and dkundel replied, "And Windows!!"
At the same time, the broader "control other devices" surface appears to have shown up before it was consistently usable. altryne said the screen was visible but did not work yet, altryne's screenshot showed a failed authorization banner, and testingcatalog's reply said the remote-control screen seemed available to only some users while Locked Use was not.
That split matters because the screenshots hint at a larger control model than simple mobile continuation. testingcatalog's screenshot advertises authorizing one Mac to control other devices on the same ChatGPT account, and the Computer use settings screenshot shows a separate Locked Use toggle for letting Codex use a Mac while it is locked. OpenAI has documented the remote connections path, but the wider device-to-device control surface is still leaking out through UI and replies rather than through a single finished announcement.