SAN FRANCISCO — OpenAI has introduced a new Computer History feature for the ChatGPT desktop app on macOS, giving users the option to let the AI system build a searchable timeline and memory of selected activity across their apps and websites.
The feature is designed to help users return to unfinished work, find information from recent tasks, and turn repeated workflows into reusable AI-assisted skills or automations. It is also likely to intensify debate over how much computer activity people should allow an AI system to observe, particularly after OpenAI disclosed that the feature’s local memory files are not encrypted.
Computer History is available to ChatGPT Pro, Business and Enterprise users on macOS. It is off by default, meaning users must actively choose to enable it. In Business and Enterprise workspaces, administrators must first authorize access before individual employees can turn it on.
The rollout does not initially include the European Economic Area, the United Kingdom or Switzerland. OpenAI says access for those regions is expected later.
For people considering whether to enable the feature, the central tradeoff is clear: Computer History may make ChatGPT substantially more useful as a contextual work assistant, but it requires careful attention to privacy settings, local-device security and the kinds of apps and websites allowed to contribute information.
What Computer History does
Computer History turns a user’s activity across approved apps and websites into a timeline and a collection of memories that ChatGPT and OpenAI’s Codex coding tool can reference.
OpenAI says users can ask natural-language questions about recent work, continue unfinished tasks, understand work habits, and convert recurring workflows into skills or automations. A user might ask ChatGPT to identify a file they reviewed earlier in the week, summarize what they worked on in a particular application, or help reconstruct steps taken during a coding project.
The feature is designed to reduce a common limitation of chatbots: lack of context. Normally, users must paste information into a chat, upload files, or repeatedly explain what they are doing. Computer History is meant to provide the AI with a controlled record of recent activity, allowing it to offer more relevant assistance.
For example, a developer who has worked across a code editor, a browser, documentation pages and a project-management tool could ask Codex to help explain a recurring issue or resume a workflow. A writer could use the timeline to locate research sources, notes, or documents connected to a recent assignment.
The system creates a stream of interaction events from the apps and websites the user permits. OpenAI says these events can include clicks, typing, keyboard shortcuts, app switches and context that macOS exposes through its accessibility system.
That does not mean Computer History records a screen video or creates a traditional keylogging file. OpenAI says the feature does not capture screenshots, microphone input, system audio or video recordings of a user’s screen.
However, the information it does collect can still be sensitive. A record of which apps a person used, websites they visited, text they typed or actions they performed may reveal details about work, health, finances, relationships, legal matters or personal communications.
How the data is processed
Computer History temporarily stores interaction events on the user’s Mac for up to 48 hours. According to OpenAI’s documentation, those events are processed on the company’s servers to generate text summaries and local memories. The event files are not retained by OpenAI after processing and are not used to train its models.
The resulting memory summaries are stored locally on the Mac as text-based files. Those memories can remain available to ChatGPT until the user deletes them.
This hybrid system differs from a feature that continuously uploads screen recordings to the cloud. It seeks to limit collection by processing selected activity events, generating summaries, and retaining the more persistent record on the user’s device.
But the privacy protection is incomplete. OpenAI explicitly warns that Computer History files can contain sensitive information, are not encrypted by Computer History and may be accessible to other programs running under the same macOS user account.
That warning is likely to be the most important consideration for security-conscious users.
An unencrypted local memory file may be readable if someone gains access to the user’s Mac account, uses the device while it is unlocked or runs malicious software under the same account. The risk is not necessarily unique to Computer History, many local files can be exposed if a device is compromised, but the feature creates a potentially rich, AI-readable summary of a person’s digital activity.
Digital Trends reported that the local files are plain-text Markdown files, making the data potentially easier to inspect if it is accessed by another application or a person using the same account.
Privacy controls users have
OpenAI has built several controls into Computer History, beginning with the fact that the feature is opt-in. It does not begin collecting activity automatically for eligible users.
Users choose which apps and websites can contribute to their history. That means a person can allow a coding environment, project-management platform or work research browser tab while excluding messaging services, banking websites, health portals, or personal email.
Users can also pause collection at any time, resume it later, inspect the generated timeline, and delete individual entries or all stored history.
To enable the feature, users open the ChatGPT desktop app on a Mac and go to Settings, then Integrations, then Computer History. They can select Turn on, review the privacy and permission details, and choose which apps and websites will be included. The feature also requires ChatGPT’s Memories setting to be enabled so it can use activity context across conversations and tasks.
In business settings, the additional administrator approval requirement is meaningful. It gives organizations an opportunity to evaluate whether the tool aligns with internal security policies, records-retention rules, client confidentiality obligations, and data-governance requirements before employees use it.
But user controls are only effective if people use them carefully. Selecting “allow all” or failing to exclude sensitive applications could allow Computer History to capture more information than a user intended.
Who should use it, and who should not
Computer History may be useful for people who work across many applications and regularly lose track of files, research, code changes, or routine processes. It could be especially appealing to software developers, researchers, writers, designers, project managers and other knowledge workers who need help maintaining context across fragmented tasks.
For a developer, it may help identify the sequence of steps that led to an error. For a researcher, it may help locate a source or document reviewed earlier. For a manager, it could turn a recurring workflow into a structured template or automation.
The feature may be less appropriate for people handling highly sensitive information, including health records, legal files, financial data, classified government materials, confidential client information or trade secrets.
Professionals who work with protected data should not assume that an opt-in consumer-style control is sufficient for regulatory compliance. They should first determine whether their employer has approved the feature and whether the organization’s security, privacy, and records-management policies permit its use.
Journalists should exercise particular caution. Activity records may reveal source identities, unpublished reporting, sensitive search terms, communications patterns, or information that could expose confidential sources. Lawyers, medical professionals, government employees, and financial advisers face similarly serious obligations.
The feature is not inherently unsuitable for such workers, but it should not be activated casually. Sensitive apps and websites should be excluded, and users should understand whether their organization has established a formal policy for AI-assisted work tracking.
Computer History versus screen recording
OpenAI’s feature has drawn comparisons to Microsoft’s Recall and OpenAI’s earlier Chronicle research preview. Those comparisons are useful because each approach seeks to solve the same problem: helping an AI system remember what a user has done on a computer.
Computer History replaces the Chronicle preview, according to reports. Unlike Chronicle and Microsoft Recall, which have been associated with screenshot-based records of user activity, Computer History is designed around interaction events and text summaries rather than screenshots.
That difference may reduce the amount of visual content captured. A screen recording can potentially include photographs, messages, passwords, video calls, documents, and content appearing briefly on a display. Interaction-event collection does not create the same kind of visual archive.
Still, the distinction should not be overstated. Clicks, typing, keyboard shortcuts, app switches and accessibility context can reveal substantial information. A text-based summary of activity may be less intrusive than a screenshot archive, but it can still create an intimate map of someone’s work and personal habits.
The real question is not whether one method is perfectly private, and the other is not. It is whether the data collected is necessary, whether users understand what is being stored, and whether they can meaningfully control and delete it.
Practical steps before enabling it
Mac users considering Computer History should take a cautious approach:
- Review the list of apps and websites that will be allowed to contribute data.
- Exclude personal messaging, email, banking, health, legal and password-management tools.
- Do not enable the feature on a shared Mac or a device without strong account security.
- Use a strong macOS password and enable device-level security protections.
- Periodically scan the timeline and memory files for unintended collection. Stop or turn off collection when processing sensitive information.
- Delete history that is no longer needed.
- Confirm whether an employer, client agreement or professional code permits the use of AI activity-tracking tools.
The most important step is to treat Computer History as a productivity feature with surveillance-like implications. It can be useful, but it changes the relationship between a user and their computer by turning ordinary work patterns into AI-readable context.
A test for AI productivity tools
Computer History reflects a broader shift in artificial intelligence: companies are moving beyond chat windows toward systems that can observe context, remember past work, and take action across applications.
That progression could make AI assistants far more useful. A system that understands a user’s recent work can save time, reduce repetitive explanations, and make assistance more relevant. But the same context that makes an AI tool valuable can also create privacy and security risks.
OpenAI’s decision to make Computer History opt-in and provide app-level exclusions, pausing and deletion controls gives users meaningful choices. Its statement that event data is deleted after processing and not used for training also addresses a major concern.learn.
However, the local memory files are still not encrypted, which is a major limitation. Before enabling the feature, users should decide whether the convenience of an AI that remembers their digital activity outweighs the risk of maintaining a local record that may contain sensitive information.
Computer History is not simply another ChatGPT feature. It is an early example of a more persistent kind of AI assistant, one designed to know not only what users ask, but what they have been doing.
