OpenAI has launched Laptop Historical past, an opt-in function within the ChatGPT desktop app on macOS that information what a consumer does throughout apps and web sites and converts that exercise into reminiscences and a searchable timeline that ChatGPT and Codex can draw on. The function appeared in OpenAIβs product changelog on August 13, 2026, and is rolling out globally to ChatGPT Professional, Enterprise, and Enterprise customers, with entry within the European Financial Space, Switzerland, and the UK to comply with within the coming weeks.
Laptop Historical past is off by default. Professional customers can flip it on themselves; in Enterprise and Enterprise workspaces, an administrator should first grant entry, and even then every member opts in individually, based on OpenAIβs documentation for the function. It requires ChatGPTβs Reminiscences system and doesn’t work with an API key or by Amazon Bedrock.
The function arrives as a rebuilt substitute for Chronicle, a analysis preview OpenAI ran earlier in 2026 that labored from display screen captures. Laptop Historical past takes a distinct technical path: it information interplay occasions uncovered by macOS accessibility options β clicks, typing, keyboard shortcuts, app switches β reasonably than pictures of the display screen.
βLaptop Historical past doesnβt depend on display screen or audio seize. As a substitute, it captures interplay occasions like clicking, typing, app switches, and extra, so it could possibly seize reminiscences quicker and extra effectively,β OpenAIβs Dominik Kundel stated within the announcement video, The New Stack reported.
How Laptop Historical past Works
The mechanics, per OpenAIβs documentation: the desktop app builds an interaction-event stream from the apps and web sites a consumer has allowed, then periodically begins an ephemeral Codex session that summarizes that stream into textual content summaries and native reminiscence recordsdata. These reminiscences are plain-text Markdown recordsdata saved on the consumerβs personal machine, beneath the Codex reminiscences listing, and customers can learn and edit them instantly.
The output surfaces in two locations. A timeline view in Settings teams exercise summaries by day and time, displaying which apps contributed to every entry and providing an choice to reveal the underlying reminiscence file in Finder or delete the merchandise. And in dialog, ChatGPT and Codex can reference the reminiscences to reply questions on current work: what the consumer was doing earlier than a break, the place a doc they seen earlier lives, a abstract of yesterdayβs work for a standup.
The extra consequential piece for OpenAIβs agent technique is what the timeline suggests. When the system identifies repeatable work within the exercise stream, a timeline entry can suggest turning that workflow right into a reusable ability or a scheduled automation, which Codex then builds from the recorded workflow. OpenAIβs documentation offers the instance of a day by day recap automation generated after the system observes the consumer assembling the identical abstract every morning.
Kundel framed the function as a change in how customers relate to the agent: βAs soon as your brokers perceive what you do and the way you’re employed, it modifications how one can interact with them,β he stated within the video, per The New Stack, asking it to create abilities from one thing simply demonstrated, or to run a standing automation, reasonably than re-explaining context every session.
The Superb Print
OpenAIβs documentation is unusually particular about what the function doesn’t gather and the place the information lives. Laptop Historical past doesn’t seize screenshots, display screen recordings, microphone enter, or system audio, and private-mode net shopping is rarely included. It doesn’t require macOSβs Display screen Recording permission. The uncooked occasion stream is saved quickly on the Mac in an remoted container, processed on OpenAIβs servers to generate reminiscences, and deleted after 48 hours; OpenAI says it doesn’t retain these occasion recordsdata after processing, until required by regulation, and doesn’t use them for coaching. The generated reminiscence recordsdata keep on the native filesystem till the consumer deletes them.
There are limits customers ought to notice in that design. The reminiscence recordsdata will not be encrypted by Laptop Historical past, and different packages working beneath the identical macOS consumer account could possibly learn them. OpenAIβs personal steerage is to guard the Mac account and exclude delicate sources. Customers can exclude or embody particular apps and web sites, pause assortment from the menu bar, and clear the final 10 minutes, final hour, final day, or all historical past, which additionally deletes the reminiscences created from it.
Two cautions within the documentation stand out for a functionality story like this one. As a result of the occasion stream can embody content material from apps and web sites, Laptop Historical past will increase publicity to immediate injection β a malicious web page might embed directions that ChatGPT or Codex would possibly later comply with. And since communication apps could be captured, OpenAI advises turning the function off throughout communications with different individuals until they’ve given prior categorical consent. Summarization additionally consumes tokens whereas it runs.
What Occurs Subsequent
Customers within the EEA, Switzerland, and the UK get entry within the coming weeks, per OpenAIβs announcement, a staged rollout that leaves the functionβs largest privacy-regulated markets for a second part. In Enterprise and Enterprise workspaces, directors management availability by a devoted permission in workspace settings, so organizational rollouts start solely the place admins grant entry.
The launch extends a reminiscence structure OpenAI has been constructing throughout ChatGPT and Codex, from saved reminiscences and chat-history reference, by the Chronicle preview, to Report & Replay, which turns a demonstrated workflow right into a reusable ability. Laptop Historical past closes the loop on that stack: the system now observes the work, remembers it, and proposes the automation itself.
