ChatGPT Computer History Remembers App Activity With 48-Hour Retention

Computer activity becomes conversational context

OpenAI has added Computer History to the ChatGPT desktop app on macOS, allowing ChatGPT and Codex to reference activity across approved apps and websites. Once enabled, the system records interaction events such as clicks, typing, keyboard shortcuts, and app switches, then summarizes them into local memories. Users can ask where they left off or turn a repeated workflow into a skill or automation.

“ChatGPT can now remember your activity across the apps and websites on your computer. With Computer History in the desktop app, future interactions feel more personalized and require less explanation.” — OpenAI

OpenAI’s account is its official channel for product, model, and deployment updates. The documentation says Computer History replaces the earlier Chronicle research preview but is a rebuilt system. Chronicle used screenshots; Computer History records interaction events and does not capture screenshots, screen recordings, microphone input, or system audio. Private-browsing activity is excluded.

Opt-in controls do not remove every risk

The feature is off by default. Pro users can enable it themselves, while Business and Enterprise administrators must first grant workspace access and each member must still opt in. Users can include or exclude specific apps and websites, pause collection from the menu bar, and delete the last 10 minutes, hour, day, or all history. It is not currently available in the EEA, United Kingdom, or Switzerland.

Raw event files are stored temporarily on the Mac and deleted after 48 hours. OpenAI processes them in an ephemeral Codex session to generate memories and says the server-side copies are not retained after processing or used for training unless legally required. The resulting Markdown memory files remain locally until the user deletes them, however, and are not separately encrypted from other software running under the same macOS account.

The documentation also warns that content encountered in apps or websites can introduce prompt-injection instructions that ChatGPT or Codex may follow. Watch for enterprise audit controls, safer defaults for sensitive apps, regional rollout, and evidence that hostile page content is reliably isolated. The source post is on X, and the Computer History documentation details permissions, retention, and privacy behavior.