OpenAI Codex adds Windows computer use and ChatGPT mobile remote control

• OpenAI shipped Windows support for Codex computer use, so Codex can now take actions directly on a Windows PC, according to OpenAI's launch post and OpenAIDevs' product summary.
• The same rollout adds remote steering from ChatGPT mobile, with OpenAI saying users can start, review, and steer work from a phone while tasks keep running on the Windows machine in OpenAI's launch post.
• OpenAI's app changelog, linked from OpenAIDevs' changelog link, pairs those Windows features with a new Codex profile that shows token stats, streaks, and longest task, which reach_vb's feature list also called out.
• The interesting backdrop is timing: two days earlier, Cua shipped Windows computer-use support for agents via MCP and CLI, and trycua's repo and blog thread explicitly named Codex as one target.

You can open the Codex changelog, skim Cua's Windows internals write-up, and compare that with francedot's background-computer-use claim that OpenAI pulled off Windows background control with "a few magic tricks." NickADobos' clip note also spotted a small UI detail, mouse traces, that makes remote actions easier to follow.

OpenAI framed the update in practical terms: Codex can now test apps, debug flows, and review work on the Windows machine where the project already lives.

That matters mostly because it closes a gap in Codex's desktop story. Until this update, OpenAI had shown computer use on other surfaces; now the Windows developer loop is part of the same product surface, with the official changelog at the Codex app entry.

The second part of the rollout is remote control from ChatGPT mobile. OpenAI's launch post says you can start, review, and steer tasks from the phone while the Windows machine keeps working, and reach_vb's feature list adds that the same control path also works from Mac.

The wording across the launch posts is consistent on one point: this is not just viewing status on mobile. OpenAI describes active steering of a running Windows host from the ChatGPT mobile app.

The quiet extra in this release is a new profile page. reach_vb's rollout list says it includes token stats, streaks, longest task, and more, while the changelog entry is the canonical source for the app update bundle.

The profile features called out in the evidence are:

• token stats
• streaks
• longest task
• unspecified additional profile fields

That is a small feature on paper, but it is the only part of the rollout that turns Codex usage into something more measurable inside the app itself.

Most of the immediate reaction was simple relief that Windows finally landed. embirico's short reaction thanked users for their patience, while koltregaskes' follow-up immediately compared the remote-control experience to Anthropic's Claude and complained about session reliability there.

There is not much deep community analysis in the evidence yet. What is there is useful anyway, because it shows where users will judge this feature first: whether remote sessions stay attached, whether background work survives app reloads, and whether Windows control actually feels dependable in day-to-day use.

Two days before OpenAI's rollout, Cua announced Windows Cua Driver support for background computer use through CLI or MCP, and trycua's launch thread explicitly pitched it for Claude Code, Codex, or custom agent loops.

Cua described three concrete pieces of context that explain why Windows support matters now:

• background computer use for agents on Windows, via CLI or MCP, per trycua's launch thread
• support for older Windows desktop software with no API, per trycua's legacy-app example
• a market of "1.4B+ monthly active devices," plus work trapped in Windows-only line-of-business apps, per trycua's market-size claim

OpenAI has not said in the evidence that Codex on Windows is built on Cua. The overlap is still notable, because francedot's background-computer-use claim described OpenAI's result as Windows background computer use with "a few magic tricks," and Cua's own Windows internals write-up exists in the same week.

A couple of late posts add details that did not appear in the launch copy. NickADobos' mouse-trace note noticed that Codex's computer-use mouse can trace paths, which is a tiny UX choice but a good one for remote review, and francedot's background-computer-use claim suggested the Windows implementation can run in the background rather than only on the visible foreground desktop.

Those are still secondary clues, not official product documentation. They are also the freshest technical hints in the evidence set, and they point to the part engineers will care about next: how OpenAI is actually brokering Windows UI control under the hood.