OpenAI Unveils Computer History Feature for ChatGPT

OpenAI has launched a new feature called Computer History, replacing its previous screenshot-based research preview known as Chronicle. Instead of capturing screen images, Computer History records interaction events on macOS, such as clicks, keystrokes, keyboard shortcuts, and application switches. It gathers this data through the operating system's accessibility system. The tool excludes screenshots, screen recordings, microphone input, system audio, and private browsing activity. These captured events are periodically summarized into text and saved locally as Markdown memory files, which are grouped by day and time in a timeline view.

For developers and professionals, this feature allows ChatGPT and Codex to access a continuous stream of context. When the system detects recurring workflows, it can suggest a "Skill" or automation, allowing users to ask Codex to generate reusable templates. Typical use cases include locating recently edited documents or automatically generating daily standup summaries. Users can control which applications and websites are tracked using include and exclude lists, and they can pause the recording at any time from the macOS menu bar. The feature is currently unavailable in the European Economic Area, Switzerland, and the United Kingdom.

Practitioners must remain cautious regarding data security and privacy. OpenAI stores temporary event files locally within the ChatGPT app group, deleting them after 48 hours. However, the generated Markdown memory files remain on the local file system in plaintext until manually deleted. Because these files are unencrypted, any program running under the same macOS user account can read them. While OpenAI processes these temporary files on its servers to generate memories, it does not retain them after processing unless legally required, and they are not used for AI training. However, subsequent chats utilizing these memories may still be used for training depending on user settings.

Furthermore, OpenAI warns of increased prompt injection risks, as malicious instructions on a webpage could trick ChatGPT or Codex if captured by the tool. The company advises against using the feature with communication apps without consent from all participants, and recommends pausing the tool or excluding apps that handle sensitive financial, health, or personal data.