OpenAI's New Feature Remembers What You Did on Your Computer — Here's How It Works
OpenAI has added a feature called Computer History to its ChatGPT desktop app for Mac. When turned on, it records what you do across your apps and websites and turns that into a timeline that ChatGPT can refer to when you ask it something. The feature was announced on August 13, 2026, and is available to people paying for ChatGPT Pro, Business, or Enterprise plans.
Computer History does not take screenshots. Instead of capturing pictures of your screen, it keeps a record of your actions — things like where you clicked, what you typed, and which apps you used. No images, videos, or audio are recorded. The idea is that when you later ask ChatGPT a question, it can look at what you were recently doing and give you a more helpful answer, suggest a shortcut, or help you finish something you started but didn't complete.
A demo video by Dominik Kundel, who works on OpenAI's Developer Experiences team, shows how it works. In the video, the app figures out the last document he edited, checks whether he shared it with coworkers on Slack, and gives him a summary of how he spent his morning. The goal is to let ChatGPT understand what you've been doing across your whole computer, not just inside the ChatGPT window.
The feature is turned off by default. Pro users can choose to turn it on. For Business and Enterprise plans, an organization's administrator can manage it. You can also tell it to skip certain apps or websites, and you can delete individual items from the timeline. Ari Weinstein, a Product and Engineering manager at OpenAI, said on X that the feature will automatically ignore anything you do in a private or incognito browser tab.
OpenAI's official X account put it simply: with Computer History, "future interactions feel more personalized and require less explanation."
The difference between this and a similar feature Microsoft tried, called Windows Recall, comes down to what gets recorded. Recall took screenshots on a regular schedule and stored them on your computer, which worried people because sensitive information — like a banking page or a private email — could end up in those images. Computer History avoids that problem by only logging text-based actions, not pictures. The downside is that a list of clicks and keystrokes is harder for a person to read through than a screenshot would be, so you have to rely more on OpenAI's own tools to show you what was captured and to delete it properly.
The controls are fairly thorough. You can turn the whole thing off, exclude specific apps or websites, and delete entries one by one. Administrators at companies can manage it for their teams, which matters because IT departments need that kind of oversight before allowing something that tracks what employees do on their computers.
The broader picture is that Computer History is part of a larger shift toward AI tools that do more than just answer questions — they're starting to act on your behalf across different programs. A model that can remember what you did this morning is a step toward one that can take over where you left off. Whether that is a good thing depends on how well it is built and how much you trust the company behind it. The design choices here — no screenshots, private tabs ignored, granular controls — are reasonable. The open question is what happens to the data over time: how long it is kept, whether it is used to train AI models, and whether deletion truly removes it. OpenAI says user actions are turned into training data. The controls are there. Living up to them is the ongoing part.
