OpenAI's Computer History turns Mac activity into searchable ChatGPT memory

OpenAI has rolled out Computer History, a new opt-in feature for the ChatGPT desktop app on macOS that records activity across permitted apps and websites, turning it into a searchable timeline and local memories for both ChatGPT and Codex. The feature requires ChatGPT Memories to be enabled and uses macOS accessibility tools to capture interaction events like clicking, typing, keyboard shortcuts, and app switching—deliberately avoiding screenshots and screen or audio recording. This replaces the earlier Chronicle research preview, which relied on screenshots, and is designed to capture memory faster and more efficiently. Users can ask ChatGPT what they were working on before a break, locate a document they last saw, or review what they accomplished during the day, with the system able to directly access the relevant file, conversation, or website from the activity timeline.

The more intriguing capability is automation: when ChatGPT spots a repeated workflow, it can suggest turning it into a skill or automation, and users can ask Codex to build a reusable workflow based on recorded activity. This pushes ChatGPT beyond simple memory into learning how users perform tasks and reproducing those steps later. Privacy controls let users choose which apps and websites contribute, pause collection from the menu bar, delete individual entries or whole time periods, and automatically exclude private-mode browsing. Temporary event files on the Mac are deleted within 48 hours after processing, but generated memory files remain locally as plain-text, unencrypted—meaning other programs under the same macOS account could potentially access them. OpenAI also warns of an increased risk of prompt injection from malicious instructions on websites or apps entering the AI's context. The feature is available to ChatGPT Pro, Business, and Enterprise users, with Business and Enterprise administrators required to enable it first. While optional and controllable, the trade-offs suggest selective use is wiser than blanket activation.