ChatGPT’s Computer History turns desktop activity into AI context

Lead

OpenAI is moving ChatGPT closer to the role of a desktop-aware assistant. According to The Verge, the macOS ChatGPT app is gaining a feature called Computer History that can turn a user’s clicks, keystrokes, and app activity into a timeline that ChatGPT and Codex can reference later.

The idea is simple but consequential: instead of relying only on what a user types into a chat box, the assistant can use recent computer activity as context. That could help it find the last document someone edited, understand whether a file was shared in Slack, summarize a morning’s work, or resume a task that was left unfinished.

Key points

• A memory layer for computer use: Computer History organizes user activity into a history that can be used by ChatGPT and Codex when responding to requests. This makes the assistant less dependent on users manually explaining what they were doing.
• Built for continuity and automation: OpenAI describes the feature as a way for ChatGPT to learn how people work, suggest automations, and pick up tasks that were left halfway done.
• Opt-in rather than default: The Verge reports that Computer History must be enabled by the user. It is not switched on automatically.
• Some controls are included: Users can exclude specific apps and websites and delete entries for more precise control. OpenAI’s Ari Weinstein also said the feature will automatically ignore content in incognito or private browser tabs.
• Not the same as Windows Recall, but close enough to invite comparison: Microsoft’s Recall drew criticism for relying heavily on screenshots. OpenAI says Computer History does not capture images, videos, or audio, and instead relies on “events.” Still, event data can reveal sensitive patterns about work, communication, and attention.

Why it matters

Computer History shows where AI assistants are heading: from conversational memory to operational memory. A chatbot that remembers your preferences is useful; an assistant that understands what you were just doing on your computer can be much more powerful. For developers, Codex could benefit from richer context around recent files, tools, and workflows. For general users, simple requests such as “continue what I was working on” or “summarize my morning” become easier to fulfill.

The trade-off is trust. Even without screenshots, a timeline of clicks, keystrokes, app usage, and document-related activity can be deeply revealing. It may not show the screen, but it can still describe behavior. OpenAI’s opt-in design, exclusion lists, deletion controls, and private browsing exception are important safeguards, yet they do not eliminate the need for users to understand what they are granting access to.

The broader industry signal is clear: the next stage of AI productivity tools will depend on access to personal context. The companies that make this feel useful, transparent, and controllable will have an advantage. The ones that make it feel like surveillance will face the same kind of backlash that has followed other attempts to give AI a memory of everything users do.

Source: The Verge AI