Back to stories
product launchGenerated by an AI editor from the reporting and web sources listed on this page.

ChatGPT on Mac Can Now Fold Your Activity Into Its Memory

OpenAI's newest feature captures clicks, keystrokes, and other Mac interactions to give ChatGPT and Codex richer context.

Published Updated The total reporting and web sources attached to this story.The AI editor’s assessment of how strongly the attached sources’ quality and agreement support this article.

What matters

  • OpenAI launched a Mac feature that lets ChatGPT use clicks, keystrokes, and other local interactions to supplement its memory.
  • The feature also feeds context to Codex, OpenAI's coding tool, potentially improving code suggestions without manual context pasting.
  • Key details—privacy controls, data retention, local vs. cloud processing, and opt-in status—remain unclear from initial reporting.
  • The move signals OpenAI's push toward OS-level assistant integration, competing with Apple's on-device intelligence strategy.

Launch facts

Platforms:
macOS

What happened

OpenAI has rolled out a new feature for ChatGPT on Mac that can incorporate your local computer activity—clicks, keystrokes, and other interactions—into the assistant's memory. According to CNET, the feature is designed to supplement context for both ChatGPT and Codex, OpenAI's coding-oriented tool, by drawing on what you do on your machine rather than relying solely on conversation history or manually uploaded files.

The feature appears to extend ChatGPT's existing memory capabilities beyond text-based recall. Instead of remembering only what you tell it in chat, the assistant can potentially use signals from your active workflow to build a more detailed picture of your tasks and preferences.

Because the source article's full body was not available at the time of this report, several specifics remain unclear: exactly which applications or interaction types are captured, whether the feature is opt-in or enabled by default, what retention or deletion controls users have, and whether the data is processed locally or sent to OpenAI's servers.

Why it matters

This is a notable shift in how ChatGPT gathers context. Memory features have so far been largely conversational—you tell the assistant something, and it remembers. Pulling in raw activity data from your operating system moves ChatGPT closer to a proactive assistant that understands not just what you say, but what you do.

For developers using Codex, the implications could be significant. If the assistant can see your recent keystrokes, file navigation, or application switches, it may be able to offer more relevant code suggestions without you having to paste context manually. That could reduce friction in coding workflows—but it also raises immediate questions about what gets captured, where it is stored, and who can access it.

For everyday users, the feature blurs the line between a chatbot and a system-level assistant. Apple's own ecosystem has been moving in a similar direction with on-device intelligence, and OpenAI's move suggests the company sees deep OS integration as a competitive frontier.

Privacy is the central tension. Keystrokes and clicks can reveal sensitive information—passwords, private messages, financial data. Without clear guardrails and granular controls, users may reasonably hesitate before granting an AI assistant that level of access. OpenAI will need to communicate precisely what is collected, how long it persists, and how users can audit or erase it.

What to watch

  • Privacy and consent details: Watch for OpenAI's documentation on whether activity capture is opt-in, what data is excluded (e.g., password fields, private browsing), and whether processing happens on-device or in the cloud.
  • Codex integration: Look for developer-facing guidance on how activity context improves Codex suggestions and whether it works across all editors or only specific environments.
  • Platform expansion: The current feature is Mac-specific. Whether a Windows or web equivalent follows will indicate how broadly OpenAI intends to pursue OS-level context.
  • Enterprise reception: IT and security teams will likely scrutinize this feature. Watch for admin controls or MDM-compatible toggles.
  • Regulatory response: Given the sensitivity of keystroke-level data, regulators in the EU and elsewhere may weigh in quickly.

What to do next

Developers

Review whether the new activity-memory feature is enabled in your ChatGPT Mac app and test its effect on Codex suggestions in your typical workflow.

If activity context improves Codex relevance, it could reduce manual context-pasting; but you need to verify what data is captured before using it on sensitive projects.

Founders

Assess whether OS-level activity context creates opportunities for your own product to integrate with or compete against ChatGPT's deeper platform access.

OpenAI's move toward OS-level context signals a market shift toward proactive assistants; founders should understand where their products fit in that landscape.

PMs

Evaluate user trust and onboarding implications if your product relies on similar activity-capture patterns, and prepare messaging around consent and transparency.

Activity-based memory raises the stakes on privacy communication; PMs should learn from how OpenAI frames (or fails to frame) its controls.

Investors

Monitor adoption signals and any regulatory or enterprise pushback to gauge whether OS-level context becomes a durable competitive moat for OpenAI.

Deep platform integration could strengthen ChatGPT's stickiness, but privacy friction could slow enterprise penetration and invite regulation.

Operators

Check whether the feature is active on team Mac devices and determine if your IT or security policy needs updated guidance on AI assistants accessing keystroke-level data.

Keystroke and click capture can expose sensitive business data; operators should act before the feature is widely enabled across the organization.

How to test

  1. 1Open the ChatGPT Mac app and navigate to settings to check whether the activity-memory feature is available and whether it is enabled or disabled by default.
  2. 2If available, enable the feature and note any consent prompts or privacy disclosures presented during activation.
  3. 3Perform a short, non-sensitive workflow—such as switching between a code editor and a browser—then ask ChatGPT a question that references your recent activity to see if it incorporates that context.
  4. 4If using Codex, open a coding task and observe whether suggestions reflect your recent file navigation or keystroke patterns without manually pasting context.
  5. 5Check for any controls to view, edit, or delete captured activity data, and test whether disabling the feature stops further capture.

Caveats

  • Full source details were not available at time of writing; exact feature mechanics, privacy controls, and availability scope are unconfirmed.
  • Avoid testing with sensitive data (passwords, private messages, financial information) until privacy boundaries are documented.
  • It is unclear whether data is processed locally or transmitted to OpenAI servers; assume cloud processing until confirmed otherwise.