OpenAI has launched Computer History, an opt-in feature in the ChatGPT desktop app on macOS that records what a user does across apps and websites and converts that activity into memories and a searchable timeline that ChatGPT and Codex can draw on. The feature appeared in OpenAI’s product changelog on August 13, 2026, and is rolling out globally to ChatGPT Pro, Business, and Enterprise users, with access in the European Economic Area, Switzerland, and the United Kingdom to follow in the coming weeks.
Computer History is off by default. Pro users can turn it on themselves; in Business and Enterprise workspaces, an administrator must first grant access, and even then each member opts in individually, according to OpenAI’s documentation for the feature. It requires ChatGPT’s Memories system and does not work with an API key or through Amazon Bedrock.
The feature arrives as a rebuilt replacement for Chronicle, a research preview OpenAI ran earlier in 2026 that worked from screen captures. Computer History takes a different technical path: it records interaction events exposed through macOS accessibility features — clicks, typing, keyboard shortcuts, app switches — rather than images of the screen.
“Computer History doesn’t rely on screen or audio capture. Instead, it captures interaction events like clicking, typing, app switches, and more, so it can capture memories faster and more efficiently,” OpenAI’s Dominik Kundel said in the announcement video, The New Stack reported.
How Computer History Works
The mechanics, per OpenAI’s documentation: the desktop app builds an interaction-event stream from the apps and websites a user has allowed, then periodically starts an ephemeral Codex session that summarizes that stream into text summaries and local memory files. Those memories are plain-text Markdown files stored on the user’s own machine, under the Codex memories directory, and users can read and edit them directly.
The output surfaces in two places. A timeline view in Settings groups activity summaries by day and time, showing which apps contributed to each entry and offering an option to reveal the underlying memory file in Finder or delete the item. And in conversation, ChatGPT and Codex can reference the memories to answer questions about recent work: what the user was doing before a break, where a document they viewed earlier lives, a summary of yesterday’s work for a standup.
The more consequential piece for OpenAI’s agent strategy is what the timeline suggests. When the system identifies repeatable work in the activity stream, a timeline entry can propose turning that workflow into a reusable skill or a scheduled automation, which Codex then builds from the recorded workflow. OpenAI’s documentation gives the example of a daily recap automation generated after the system observes the user assembling the same summary each morning.
Kundel framed the feature as a change in how users relate to the agent: “Once your agents understand what you do and how you work, it changes how you can engage with them,” he said in the video, per The New Stack, asking it to create skills from something just demonstrated, or to run a standing automation, rather than re-explaining context each session.
The Fine Print
OpenAI’s documentation is unusually specific about what the feature does not collect and where the data lives. Computer History does not capture screenshots, screen recordings, microphone input, or system audio, and private-mode web browsing is never included. It does not require macOS’s Screen Recording permission. The raw event stream is stored temporarily on the Mac in an isolated container, processed on OpenAI’s servers to generate memories, and deleted after 48 hours; OpenAI says it does not retain those event files after processing, unless required by law, and does not use them for training. The generated memory files stay on the local filesystem until the user deletes them.
There are limits users should note in that design. The memory files are not encrypted by Computer History, and other programs running under the same macOS user account may be able to read them. OpenAI’s own guidance is to protect the Mac account and exclude sensitive sources. Users can exclude or include specific apps and websites, pause collection from the menu bar, and clear the last 10 minutes, last hour, last day, or all history, which also deletes the memories created from it.
Two cautions in the documentation stand out for a capability story like this one. Because the event stream can include content from apps and websites, Computer History increases exposure to prompt injection — a malicious page could embed instructions that ChatGPT or Codex might later follow. And because communication apps can be captured, OpenAI advises turning the feature off during communications with other people unless they have given prior express consent. Summarization also consumes tokens while it runs.
What Happens Next
Users in the EEA, Switzerland, and the United Kingdom get access in the coming weeks, per OpenAI’s announcement, a staged rollout that leaves the feature’s largest privacy-regulated markets for a second phase. In Business and Enterprise workspaces, administrators control availability through a dedicated permission in workspace settings, so organizational rollouts begin only where admins grant access.
The launch extends a memory architecture OpenAI has been building across ChatGPT and Codex, from saved memories and chat-history reference, through the Chronicle preview, to Record & Replay, which turns a demonstrated workflow into a reusable skill. Computer History closes the loop on that stack: the system now observes the work, remembers it, and proposes the automation itself.
Credit: Source link
