ChatGPT's new 'Computer History' turns everyday work into AI context

OpenAI has introduced Computer History for ChatGPT, a feature that records user activities across apps and websites, turning them into a searchable timeline with memories that ChatGPT and Codex can use as context. Unlike the earlier 'Chronicle' research preview which relied on screenshots, Computer History captures 'interaction events' such as clicks, keyboard inputs, shortcuts, and app switches, read via macOS accessibility, without screenshots, screen recordings, microphone input, or system audio, and it never includes private browsing activity. The captured events are periodically summarized into text and stored locally as Markdown memory files; the timeline groups summaries by day and time, showing involved apps. A key element is workflow recognition: when ChatGPT detects repeatable patterns, it suggests a 'skill' or automation, and users can ask Codex to create reusable templates from recorded workflows, such as finding recently edited documents or creating standup summaries. Access is multi-level: for business and enterprise workspaces, an administrator must enable it, and each user must individually consent, with the Memories feature required. Users can control which apps and websites contribute via include/exclude lists, pause/resume via the macOS menu bar, and the feature is currently unavailable in the European Economic Area, Switzerland, and the UK. Temporary event files are stored locally in the ChatGPT app group and deleted after 48 hours, while memory files remain as plaintext Markdown until manually deleted, without encryption, so other programs under the same macOS account could access them. OpenAI processes temporary event files on its servers to generate memories but says it does not retain them unless legally required, and the files are not used for AI training, though chat contents may become training material if memories are used as context in later chats per user data control settings. OpenAI warns of elevated prompt-injection risk: malicious instructions on a website could lead ChatGPT or Codex to follow them, advises not to use the feature with communication apps without consent, and recommends pausing or excluding apps with sensitive health, financial, or personal data.