OpenAI released Computer History, a Mac application that logs every click, keystroke, and app switch on a user's device, then converts that activity into a searchable timeline accessible to ChatGPT and Codex. The tool turns raw computer activity into machine-readable data that AI assistants can reference during conversations.
The feature stores data locally as unencrypted Markdown files on the user's machine. OpenAI states the logged activity itself is not used directly for AI model training. However, the company acknowledges that memories generated from this data, which feed into ChatGPT conversations, may eventually become training data through normal usage patterns.
This represents a shift in how AI assistants access context about user behavior. Rather than requiring manual prompts or file uploads, Computer History creates an always-on record of device activity. Users can query ChatGPT about what they did at specific times, retrieve files they worked on, or get AI assistance reconstructing their work history.
The unencrypted local storage approach carries privacy implications. Data lives on the user's machine by default, but remains vulnerable if the device is compromised or physically accessed. OpenAI hasn't detailed encryption options or whether encrypted storage will become available. Users who sync devices or use multiple computers face additional choices about data scope.
The distinction between "not used for training" and "memories may become training data" matters. OpenAI separates the raw activity logs from the processed memories ChatGPT generates in conversations. The logs themselves stay local. But when users ask ChatGPT questions about their activity, and ChatGPT references those memories in responses, that conversation content follows standard ChatGPT terms, which permit usage for model improvement unless explicitly opted out through privacy settings.
This creates a gray zone. A user's keystroke log stays private. But ChatGPT's summaries of that activity, discussed in conversation, become fair game for training unless the user specifically disables chat history. OpenAI has not clarified whether Computer History memories inherit the same privacy toggles as regular ChatGPT conversations, or if they operate under different rules.
The technical implementation reveals trade-offs. Comprehensive logging enables powerful recall and context-aware AI assistance. It also generates detailed behavioral data that, even stored locally, represents a concentration of sensitive information. A stolen laptop becomes a window into months of work patterns, visited sites, documents edited, and code written.
OpenAI's approach differs from other OS-level monitoring tools. Apple's Spotlight search indexes files locally for fast searching. Windows Search offers similar functionality. But Computer History explicitly pipes that data to an AI model trained on internet-scale datasets. The implications extend beyond simple file search.
Early adopter response will shape how this feature evolves. If privacy concerns dominate, OpenAI may add encryption, adjust training data policies, or limit scope. If adoption proves heavy, expect similar features from competitors. Google, Microsoft, and Anthropic all work on AI assistants that need better context about user activity.
Computer History launches as OpenAI expands beyond chat interfaces into desktop integration. The feature positions ChatGPT as a system-level tool rather than a browser tab. That shift requires trust that activity data stays protected and usage terms remain transparent. OpenAI's current disclosure leaves both questions partially answered.