ChatGPT’s Computer History tracks your clicks and keystrokes

OpenAI has released an update to the ChatGPT desktop client for macOS that introduces a feature called Computer History. The tool silently records mouse clicks, keyboard presses, and application switches while you work, then turns that raw activity into a chronological timeline. ChatGPT and the related Codex model can query this timeline to suggest next steps, automate repetitive actions, or resume tasks you left unfinished. The data is stored locally and used to improve the model’s understanding of your personal workflow.

What You Need to Know

The Computer History feature is enabled by default when you install the latest macOS build of the ChatGPT app. It operates in the background, capturing low‑level input events such as which window is active, the timing of keystrokes, and the coordinates of mouse clicks. Those events are aggregated into a timeline that groups actions by application and by user‑defined projects, allowing the model to see patterns like “you often open a spreadsheet after checking email” or “you tend to leave a draft document open for several hours.”

When you ask ChatGPT for help—whether it’s drafting a reply, generating code, or summarizing a document—the model can reference your recent timeline to infer context. For example, if you have been editing a Python script and then ask for a function suggestion, the model can look at the last few minutes of activity to see which libraries you imported and what variable names you used, producing a more relevant answer. The feature also powers a new “suggest automation” button that proposes a simple AppleScript or Shortcut based on repeated sequences it detects.

Why It Matters

By turning everyday computer use into training data, the feature blurs the line between passive assistance and active workflow integration. Users who rely on repetitive tasks—such as data entry, code refactoring, or report formatting—may see tangible time savings as the model learns to anticipate the next step and offer ready‑to‑run snippets. This approach also raises questions about data privacy, since the timeline contains potentially sensitive information like passwords typed in fields or confidential document contents.

From a product perspective, the move signals OpenAI’s effort to make its models more than just chatbots; they are becoming context‑aware agents that can observe and adapt to a user’s environment. For developers who already use Codex for code generation, the added timeline context could reduce the need to repeatedly explain project specifics, streamlining the collaboration between human and machine.

Key Details

• Computer History records clicks, keystrokes, and window focus changes on macOS.
• Data is stored locally on the user’s machine and is not uploaded unless the user opts in to share improvement data.
• The timeline groups actions by application and can be segmented into custom projects.
• ChatGPT uses the timeline to provide more context‑aware responses and to suggest automations.
• Codex can reference the same timeline when generating code snippets, reducing repetitive explanations.
• Users can disable the feature or clear the history from the app’s Settings pane.

What’s Next

OpenAI plans to roll out similar activity‑based context features to the Windows and Linux versions of the ChatGPT desktop app later this year, with additional controls for granular data retention. Future updates may allow users to define which applications are monitored and to export their timeline for personal analysis or backup. Feedback from the macOS beta will shape how aggressively the model proposes automations and how transparent the data usage disclosures become.

📌 Source: Verge Ai