Codex updates app with customizable shortcuts and 10-50x faster Git ops

• OpenAI pushed a fast follow set of Codex app fixes, with OpenAIDevs' shortcuts post adding customizable keyboard shortcuts and moving core Git actions back into the review flow.
• OpenAIDevs' performance breakdown says the Codex app now does about 75% less re-rendering on thread switches, while expensive Git operations in large repos got 10x to 50x faster.
• The UI cleanup also touched the thread panel update and local server list cleanup, both aimed at reducing friction in longer coding sessions.
• OpenAI separately reset Codex usage caps across paid plans, according to thsottiaux's reset announcement, though thsottiaux's follow-up said some accounts saw delayed propagation.
• The reset landed a day after thsottiaux's incident update said the team had fixed two issues that could explain reports of degraded GPT-5.5 capability inside Codex.

The weekend ship bundled a few different Codex storylines together. You can open the remote connections docs from thsottiaux's mobile-control reply, watch OpenAIDevs' demo thread walk through shortcut and review-flow changes, and check testingcatalog's settings screenshot for an unreleased "Locked use" toggle that would let Codex keep using a Mac while it is locked.

The most obvious changes are small, practical ones. OpenAIDevs says shortcuts are now customizable from settings, and the same thread says common Git actions were moved back into the review flow.

That review-flow list is concrete:

• commit
• push
• branch
• PR creation
• PR status

This looks like OpenAI closing paper cuts instead of chasing a big redesign. dkundel's reply called it "Lots of little paper cuts addressed," and thsottiaux's weekend note said the team shipped immediately because the changes made Codex "a lot more delightful to use."

The thread header now carries more of the session state. According to OpenAIDevs' thread panel post, summaries, local state, Git context, sources, and related controls now load from the header more cleanly.

The local server list got its own cleanup pass:

• better filtering
• remembered sort state
• clearer empty states
• connected route state
• refreshes for non-listed ports every 120 seconds

Those are the kind of changes that matter more in real repos than in launch demos, especially if Codex is acting as a local tool hub instead of a single chat window.

OpenAI attached actual numbers to this part of the ship. OpenAIDevs' performance post lists five changes:

• about 75% less re-rendering when switching threads
• some streaming paths cut to zero unnecessary re-renders
• expensive Git operations in large repos reduced by 10x to 50x, depending on the operation
• less UI churn across streaming responses, thread switching, and sidebar interactions
• faster time to usefulness around startup and first interaction

The large-repo Git claim is the standout. Most UI polish posts stop at "feels faster." This one names the expensive path, large-repo Git operations, and puts a wide but still material improvement range on it.

OpenAI reset Codex usage limits across all paid plans on Friday. In the same thread, thsottiaux's remote-control reply pointed users toward the new ChatGPT mobile remote-control feature.

The reset was not perfectly clean. thsottiaux's follow-up said some accounts were seeing propagation issues and that the team was looking into it.

That came shortly after a separate reliability thread. thsottiaux's earlier incident post said the team was investigating reports that GPT-5.5 was performing worse for some Codex users, and thsottiaux's later update said OpenAI had found and fixed two issues that could explain capability degradation over roughly the prior 48 hours, with another usage reset promised that evening.

One user comparison from haider1's usage-limit observation also helps explain why the reset mattered: they noted that Codex keeps working after you hit 100% usage if an execution plan is already in flight.

The app fixes landed in the middle of a broader Codex push onto phones. ChatGPTapp's preview post said Codex was rolling out in preview inside the ChatGPT mobile app, and reach_vb's reposted meetup claim said the product had already grown past 4 million weekly active users.

The next layer may be more aggressive remote control. testingcatalog's screenshot shows a settings entry for "Locked use," described as "Let Codex use your Mac while it's locked," while kimmonismus' follow-up tied that setting to OpenAI's half-open-laptop teaser photo from the prior day.

Support also appears to be broader than the original Mac framing. koltregaskes' Windows claim said Codex remote control works on Windows, and dkundel's reply answered simply, "And Windows!!" That lines up with the existence of the remote connections docs, even if the locked-screen mode itself still looks pre-release.