OpenAI Codex supports background computer use with Mac app control and Telegram BotFather setup

• OpenAIDevs' demo post showed Codex using "computer use" to work across Mac apps in the background instead of taking over the active desktop.
• Early hands-on reports from steipete's Telegram setup and steipete's Google Cloud example pushed that beyond repo edits into authenticated desktop flows, including BotFather token setup and clicking through cloud admin screens.
• The Codex app's in-app browser is turning into a lightweight test harness, according to thsottiaux's product note, OpenAIDevs' repost, and TheRealAdamG's retweet of pranaveight.
• Long-running workflows are part of the same shift: TheRealAdamG confirmed /goal is now in the app, while reach_vb's automation example showed scheduled thread automations that wake back up in the same context.

You can watch OpenAI's own computer use walkthrough, browse the Peekaboo vision tool that one early user used to inspect Telegram windows, and check a separate OpenClaw proof-generation example for the kind of UI-testing loop people are already building around Codex. There is also a leaked-looking Codex mobile setup screen and a report from btibor91's rollout summary that the Chrome extension was available outside the EU and UK.

OpenAI's framing in OpenAIDevs' post was narrow but important: Codex can click, type, and move across apps while the user keeps using the Mac. sound4movement's hands-on note said the earlier version felt like "two people at the same machine," while the background version felt smooth enough to leave running.

The interesting early examples were authenticated, stateful tasks. In steipete's Telegram post, Codex opened the Telegram Mac app, talked to BotFather, and tried to fetch a bot token using Peekaboo for screen inspection. In steipete's Google Cloud post, it noticed an API was disabled and started clicking through Google Cloud Admin to enable it.

The app browser keeps picking up features that make it look less like a preview pane and more like a QA surface. thsottiaux listed different viewports, screenshot viewing, better annotations, and token-efficiency improvements, while OpenAIDevs' repost highlighted viewport-size testing directly.

Users are already treating it that way. kevinkern's post said Codex was adding features and running automated UI tests in the iOS simulator, and kevinkern's serve-sim note described a helper that captures the simulator inside the in-app browser. TheRealAdamG's retweet of pranaveight boiled the whole thing down neatly: the Codex app IAB is basically an automated front-end test harness.

The other pattern in the evidence is persistence. TheRealAdamG said /goal has moved into the Codex app, not just the CLI.

That shows up in three distinct ways:

• reach_vb's automation example showed thread automations that wake Codex on a schedule inside the same thread, preserving context instead of starting fresh.
• doodlestein's react-doctor run said a fully autonomous /goal pass spent around 15 hours pushing a React app to a score of 100.
• doodlestein's usage-limit post showed 11 separate /goal runs still chugging along after GPT Pro limits were hit.

Multiple posts surfaced the same "Set up Codex mobile" screen. WesRoth's screenshot and koltregaskes' screenshot both described a ChatGPT app flow for accessing threads and projects from a phone and getting notified when Codex desktop needs attention.

That mobile hint sits next to a messy rollout map. btibor91's summary said the Codex Chrome extension runs on macOS and Windows and was available everywhere except the EU and UK, while kimmonismus' complaint separately said computer use was not available in Europe.