OpenAI's New Computer History Feature Turns Your Mac Clicks and Keystrokes Into ChatGPT Memory

SAN FRANCISCO — OpenAI has introduced a new feature for the ChatGPT desktop app on macOS called Computer History that records a user's clicks, keystrokes and app activity to build a searchable timeline the AI assistant can reference when helping with tasks, a tool that has drawn comparisons to Microsoft's controversial Windows Recall feature.

Computer History appeared in OpenAI's desktop app changelog on Aug. 13 and is described by the company as an opt-in tool that tracks user activity across approved apps and websites, converting that activity into structured memories and a timeline that both ChatGPT and OpenAI's coding tool Codex can draw on when a user makes a request. The feature is off by default and must be actively enabled by users.

The tool works by capturing interaction events exposed through macOS's built-in accessibility system, including mouse clicks, typing, keyboard shortcuts and switches between applications. Unlike Chronicle, the earlier research preview the feature replaces, Computer History does not rely on screenshots, video or audio recordings. Dominik Kundel, a member of OpenAI's Developer Experiences team, explained the distinction in a demonstration video accompanying the launch. "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," Kundel said.

In the same demo video, Kundel showed the assistant looking up the last document he had edited, checking whether it had been shared with colleagues through Slack, and generating a summary of how he had spent his morning, illustrating the kind of task-recall functionality OpenAI is positioning the feature to support.

Ari Weinstein, a product and engineering manager at OpenAI, addressed privacy safeguards built into the tool in comments posted on the social platform X. Weinstein said Computer History will automatically exclude content viewed in incognito or private browsing tabs, and that users retain the ability to exclude specific apps and websites from being tracked altogether, as well as delete individual entries from their history after the fact.

According to OpenAI's own documentation, Computer History periodically converts the raw stream of interaction events into text summaries and local memory files, which are then organized into a timeline grouped by day and time. Each entry in the timeline includes a title, a text summary, and a list of the apps that contributed to that particular activity record. The system is designed to help users resume half-finished tasks, locate recently accessed files, generate summaries of completed work, or identify repetitive processes that could be converted into reusable automations or skills.

The feature is currently available to ChatGPT Pro, Business and Enterprise subscribers on macOS, according to multiple outlets that have reviewed the rollout. Access for users in the European Economic Area, Switzerland and the United Kingdom has not yet begun and is expected to follow in the coming weeks, reflecting the additional regulatory considerations OpenAI faces in those markets regarding data collection and user privacy.

For Business and Enterprise workspace accounts, enabling Computer History requires an additional layer of approval. An administrator must first grant access to the feature at the organizational level before individual employees are able to choose whether to turn it on for their own accounts, giving workplace administrators a degree of oversight over whether the tool can be used at all within their organization.

OpenAI has said that data collected through Computer History is not used to train its underlying AI models. The company has also advised users to exclude sensitive applications from tracking and to pause the feature during communications with other people who have not given explicit consent to have their conversations captured, given that messaging and communication apps can fall within the scope of what the tool records.

The launch has inevitably drawn comparisons to Windows Recall, a Microsoft feature introduced in 2024 that similarly built a searchable record of user activity on Windows PCs. Recall faced substantial backlash following its announcement due to security and privacy concerns tied to its reliance on continuous screenshots of a user's screen, which security researchers warned could create a significant new target for malicious actors seeking to extract sensitive information. Computer History's design, which avoids capturing screen images entirely in favor of structured interaction events, represents a different technical approach, though it raises some of the same fundamental questions about the tradeoffs between AI-assisted convenience and continuous personal activity monitoring.

Security researchers examining the feature have also raised concerns about its underlying technical implementation. According to one analysis, the memory files Computer History generates on a user's Mac are not encrypted, meaning other programs running under the same macOS user account could potentially read them. The feature also relies on macOS's accessibility permission system, the same system-level access that security researchers have long identified as a common target exploited by keylogging and stalkerware malware, though Computer History reads text a user has already typed rather than intercepting keystrokes in real time as they are entered.

The rollout of Computer History comes roughly a month after Codex, OpenAI's AI coding assistant, was integrated directly into the ChatGPT desktop application, part of a broader push by the company to expand ChatGPT's capabilities as a persistent assistant capable of understanding ongoing work across a user's computer rather than requiring users to repeatedly re-explain context in each new conversation. OpenAI has also cautioned that the deeper level of system access required by Computer History introduces an increased risk of prompt injection, a type of security vulnerability in which malicious content encountered in an app or website could potentially manipulate the AI assistant's behavior, underscoring the tradeoffs the company is asking users to weigh in exchange for the feature's expanded contextual awareness.