OpenAI wants ChatGPT to remember your workday better than you do. The newly introduced Computer History feature, currently rolling out in the ChatGPT desktop app for macOS, turns your clicks, keystrokes, web browsing and more into a searchable timeline.
The idea is to make ChatGPT more of an intelligent assistant. It can help you resume unfinished work, locate relevant files or conversations and automate workflows. If you lose the thread of what you were doing before lunch, ChatGPT can reconstruct it for you. Whether you want an AI logging all that activity is a separate question — one that’s in opposition to Apple’s privacy positioning.
What ChatGPT Computer History actually does
As agentic AI takes off, all the big AI companies are trying to transform their offerings from passive chatbots into active AI agents that can get things done. The new Computer History feature is a step toward agentic AI that’s deeply embedded in the user’s personal context and workflow. It provides ChatGPT with a “provenance layer”: a record of what the user was doing, where they did it, and how they did it. This context should make ChatGPT’s assistance more proactive and automations more personalized.
Computer History replaces Chronicle, the research preview OpenAI launched for Codex back in April. Chronicle worked by taking screenshots, but Computer History throws out that approach.
Instead, it records interaction events and does not capture your screen or audio. Clicks, types, keyboard shortcuts, and app switches are captured using macOS’s accessibility system instead of a camera pointed at your screen.
Periodically, ChatGPT turns that event stream into plain-text summaries and local memory files. It also builds a timeline you can browse in Settings. Each entry shows what you worked on, and which apps were involved.
Sometimes it also suggests a skill or automation if it spots a repeatable workflow — like turning your weekly launch-update process into a one-click task.
And once it’s on, you can ask ChatGPT what you were working on before your last break, or where that proposal document went. It searches your recent activity and points you in the right direction. No more reconstructing your morning from memory.
Opt-in, but not simple
Computer History is turned off by default. But you can turn it on by heading over to Settings > Integrations, selecting Computer History, and following the prompts. The feature is currently available to Pro, Business, and Enterprise ChatGPT users.
It also requires ChatGPT’s Memories feature to be turned on, since that’s what lets the timeline persist across chats. For now, it’s available in the U.S. — but not available in the EU, Switzerland, or the UK. That’s a familiar asterisk for any feature touching your personal data.
You do get control once it’s on. Pause collection from the Mac menu bar, exclude specific apps or websites. You can wipe the last 10 minutes, or your entire history. Also, Private browsing is excluded automatically.
How this stacks up against Apple’s approach
Apple has built its pitch around keeping this kind of data on your device. Apple Intelligence’s on-device processing and Private Cloud Compute are both designed so Apple itself can’t see what you’re doing.
Computer History takes the opposite bet: it processes your activity on OpenAI’s servers, even temporarily, to generate those memory summaries.
But that doesn’t mean Computer History is reckless. The controls are real, and OpenAI is upfront about its tradeoffs. This gives Mac users a real choice between two different philosophies of AI assistance, sitting side by side on the same machine.
The privacy fine print you should actually read
OpenAI says it possesses your activity data on its servers to generate memory summaries. But the raw files are only deleted after 48 hours, with ChatGPT not using them for training purposes.
Then there’s a red flag as well. OpenAI says Computer History raises the risk of prompt injection from malicious content in the apps and websites you use.
This is something worth remembering the next time you’re tempted to grant an AI assistant a standing window into your screen. OpenAI is betting people will trade some of that access for convenience. And that bet paying off may say more about AI companies’ trustworthiness than about the feature itself.
