ChatGPT can now remember what you do outside the chat: how Computer History works

You open a spreadsheet, fix two numbers, jump into email to reply to a client and finish the morning preparing a website. In the afternoon you ask ChatGPT for help and have to explain the whole job again. That tax is what OpenAI wants to remove with Computer History, a new desktop feature that remembers activity you choose to share from other apps and websites.

Until now, ChatGPT's memory began with what you told it in conversation. It can now rebuild context from work that happened outside the chat. The distinction sounds small until you ask, 'Where did I leave yesterday's sales comparison?' and the assistant already knows.

From 'here is the context' to 'I know where you left off'

Computer History builds a searchable timeline from interactions such as switching apps, typing, clicking and using shortcuts, together with accessible context from the active window. It does not save a film of your screen or listen through the microphone. Instead, it turns those interactions into memories that ChatGPT and Codex can retrieve later.

OpenAI's example shows something more useful than a list of recently opened apps. ChatGPT remembers that you reviewed sales-tracker exceptions in the morning, then mapped launch priorities and, after spotting a repeated routine, suggests turning it into an automation. That is the play: ChatGPT Work could already sign into your accounts and continue a task; now it wants to arrive knowing what happened before you called it.

For anyone working across documents, email, a browser and several tools, that could remove the most tiresome part of using an AI: writing a briefing so the machine understands the project. In Codex, it has another obvious use. A repeated sequence can become a suggested skill or automation instead of tomorrow's manual chore.

It does not record your screen, but this is still a major permission

OpenAI has made Computer History opt-in. You can choose which apps and websites feed it, pause collection, clear the history or disable it completely. Business and Enterprise administrators must enable the feature before individual users can opt in. Those controls matter because an orderly work history may reveal more than a stray screenshot: clients, schedules, projects, habits and the tasks you keep postponing.

Our advice is simple: enable it by project, not for the entire computer. An editor may grant access to the browser and publication folder without bringing WhatsApp, banking or health into the same memory. It is also worth checking ChatGPT Memory and 'Improve the model for everyone' first, as OpenAI says your account's training controls apply to Codex too. Convenience, yes. A blank cheque, no.

Europe will have to watch from the outside for now

The first rollout is limited to the ChatGPT desktop app on macOS and Pro, Business and Enterprise plans. It also excludes the European Economic Area, the United Kingdom and Switzerland for now. OpenAI has not given a date for wider availability.

That is frustrating for people working from Europe, although this time the delay may be useful. I would rather watch this memory deal with months of real work before handing it a complete diary of my computer. If it performs as promised, the ChatGPT that predicts the next word will have gained something more valuable than another benchmark: continuity.