OpenAI Computer History Turns Your Mac Activity Into ChatGPT Memories — No Screenshots Required

OpenAI positions this as a privacy-forward choice: "Computer History turns your activity across apps and websites into memories and a timeline without screenshots or recordings" . But The Register noted that the approach effectively replaces screenshot surveillance with "friendly keylogging" — all typed input in selected apps becomes part of the record .

Who Can Use Computer History

Regional access is expected to expand: OpenAI stated that access in the EEA, UK, and Switzerland will follow in the coming weeks .

How Raw Interaction Events Are Stored and Processed

The data pipeline works in six stages, striking a balance between local privacy and server-side intelligence :

1. User selects sources — after opt-in, you choose which apps and websites to include or exclude .
2. Local event recording — interaction events are captured on the Mac inside the ChatGPT App Group sandbox .
3. Temporary local storage — raw event files stay on the Mac for up to 48 hours .
4. Server-side summarization — an ephemeral Codex session periodically processes the event stream on OpenAI's servers to generate text summaries .
5. Memories returned locally — the output is stored as plain-text Markdown memory files and human-readable timeline entries on the Mac .
6. Later retrieval — ChatGPT or Codex can reference those memories for context .

OpenAI says the new implementation reduces token usage compared to Chronicle, but did not quantify the reduction in its announcement .

Privacy Controls and Risks

Controls Available to Users

From Settings → Integrations, users can :

• Inspect timeline entries
• Reveal memory files in Finder
• Delete specific items
• Clear the last 10 minutes, hour, day, or all history
• Pause collection at any time
• Manage which apps and websites contribute

Key Risks

Prompt injection warning: OpenAI's own help center warns that when agents access sensitive data — including context from Computer History — they create potential privacy risks including "prompt injection" attacks, though ChatGPT agent incorporates multiple mitigation layers . Prompt injection is a well-documented attack vector where malicious instructions hidden in web pages or content sources can hijack AI agents . In February 2026, OpenAI introduced Lockdown Mode specifically to mitigate these risks .

No encryption: Computer History files are not encrypted by the feature itself, and OpenAI warns that "other programs running as your macOS user may be able to access them" . This means any app running on your Mac with your user privileges could potentially read the plain-text memory files.

Sensitive data exposure: While screenshots and audio are excluded, typing and accessibility context can expose passwords, confidential messages, customer information, health data, financial records, and other sensitive material captured during normal work . OpenAI notes that "Computer History files can contain sensitive information" .

Automated Workflow Capabilities

Computer History is designed to identify repeated workflows and prompt users to turn them into reusable skills or automations . The timeline and memories feed into both ChatGPT and Codex, enabling :

• Resuming interrupted work — asking "What was I doing before break?" to pick up where you left off
• Finding recent documents — locating a file you viewed earlier but can't remember the name
• Generating daily summaries — a recap of the day's activity across apps
• Skill creation — ChatGPT may suggest converting routine multi-step tasks into a reusable skill or automation that Codex can execute
• Codex automations — the same event context is shared with Codex, allowing it to chain together actions across apps with minimal re-prompting

For example, if you regularly download, convert, and upload files in a specific workflow, Computer History might recognize the pattern and offer to automate it as a Codex skill — reducing the need to repeatedly describe the process.

Bottom Line

Computer History represents OpenAI's attempt to offer persistent context for ChatGPT without the privacy backlash that Microsoft's Recall feature generated. By swapping screenshots for interaction events and keeping raw data local, OpenAI gives users more granular control — but the feature still comes with real privacy risks, including prompt injection vulnerabilities and unencrypted storage of sensitive typed content. For Pro, Business, and Enterprise users willing to trust these safeguards, the payoff is a ChatGPT that remembers what you've been doing across all your apps, not just inside the chat window.