OpenAI shipped a feature this week that turns your Mac into ChatGPT’s memory.
Computer History is an optional tool in the ChatGPT desktop app on macOS that records activity across apps and websites, then lets ChatGPT and Codex reference it so you can pick work back up without re-explaining everything.
It is a big deal for anyone living in ChatGPT, and it is also exactly the kind of feature that deserves a careful read before you switch it on.
What computer history actually records
The pitch is simple. Instead of describing what you were working on, you ask ChatGPT, and the app already knows. OpenAI says the feature turns activity across apps and websites into memories and a timeline that ChatGPT and Codex can use. Choose which apps and websites contribute, pause collection, and review or delete your history at any time.
The technical choice matters more than the marketing. Computer History records interaction events, not screenshots.
It captures clicks, typing, keyboard shortcuts, and app switches through macOS accessibility features, which is a completely different approach from the old Chronicle research preview. Chronicle relied on screen captures. OpenAI says the rebuilt system is faster and cheaper to run, and it avoids the creepiest part of the earlier design.
The release notes put the boundaries in writing. Computer History does not capture screenshots, screen recordings, microphone input, or system audio, and private browsing is excluded.
The raw interaction events stay on your Mac temporarily, get processed to generate memories, and are not retained or used for training. The plain-text memory files remain stored locally until you delete them.
Who gets it, and how it works
Computer History is off by default. Pro users can turn it on themselves, while Business and Enterprise workspaces require an administrator to grant access before members can opt in. OpenAI is not offering it to free users, and it is not currently available in the EEA, UK, or Switzerland.
Once it is on, you get a searchable timeline in Settings. You can inspect entries, reveal individual memory files in Finder, delete specific items, or clear activity from the last 10 minutes, hour, day, or all time.
OpenAI says the timeline can help you figure out what you were doing before a break, find a recently viewed document, or generate a summary of the day’s work. It may also suggest turning repeated workflows into skills or automations, which is where this starts to feel like an agent feature rather than a memory feature.
That agent angle is why I care.
The whole point of the agent push right now is software that understands context the way a human coworker does, and the escapes and containment probes I have covered keep showing how much context matters.
A tool that quietly builds a model of your working day is a tool that makes the next round of agents much more useful. It also makes them much more dangerous if something goes wrong.
The privacy honest take
Here is the part OpenAI put in the docs that most coverage will skip: Computer History can increase prompt-injection risk from malicious content in apps or websites.
If a webpage you visited contains instructions designed to manipulate the model, that content can now influence future responses through your history. OpenAI recommends pausing collection during communications with other people without their express consent, and excluding apps with sensitive information.
That is the right warning, and it deserves more attention than the feature itself. Microsoft’s Recall burned the company because it was the same idea without the guardrails.
Computer History is the careful version: opt-in, off by default, no screenshots, local files, explicit limits. The privacy controls are genuinely good, and the data boundaries are written down in plain language.
None of that changes the core tradeoff. You are handing OpenAI a record of how you actually spend your day, and even with good controls, that record is valuable and sensitive.
The prompt-injection warning is OpenAI admitting the model can be tricked into misusing what it learns. I wrote about OpenAI’s price cuts earlier this summer, and the pattern is consistent: OpenAI keeps shipping memory features that make the product more useful while asking users to trust the safeguards.
Should you turn it on
If you are a Pro user who lives in ChatGPT and Codex, Computer History is genuinely useful. The ability to ask what you were working on yesterday and get a real answer is a workflow upgrade, and the implementation is noticeably more careful than anything Microsoft shipped.
If you work with sensitive documents, talk to people who have not consented to being recorded, or just do not want your app usage summarized into someone else’s model, leave it off. The feature being optional is the whole point.
This is the version of AI memory done right, and I would expect the rest of the industry to copy it within a year. Just read the prompt-injection warning first.
