OpenAI
OpenAIAug 13
Tech

Computer History in ChatGPT

3 min video3 key momentsWatch original
TL;DR

OpenAI's Computer History feature lets ChatGPT and Codex watch your desktop activity to understand context without screenshots, enabling more natural task automation and personalized workflows.

Key Insights

1

Interaction events, not screenshotsComputer History captures interaction events like clicks and typing, not screen recordings—making it faster and more efficient while giving granular privacy controls per app or website.

2

Suggests automations from patternsThe agent can suggest automations based on repetitive work patterns you do, then create skills to run those tasks automatically—like drafting standup updates every morning.

3

Memories stay localAll captured memories live locally on your file system so you can inspect, delete, or clear entries individually without relying on cloud storage.

Want this for every new video OpenAI posts? Brevyd summarizes each upload automatically, the morning it drops.

Deep Dive

How Computer History Works

Computer History is a new macOS feature that records what you do on your desktop—window switches, clicks, typing—and converts it into a timeline the AI can reference. Unlike screen recordings, it captures only interaction events, making the system faster and more efficient. Once enabled in settings, ChatGPT and Codex remember your recent activities. In the demo, when asked about the last Google Doc edited, the agent immediately retrieved it from the captured history. This context persistence lets users ask followup questions naturally without restating the full situation each time.

Privacy and Control

OpenAI built granular controls directly into the feature. Users can quickly exclude entire websites or apps from the menu bar, or configure blocklists in settings. All captured memories remain on the local file system—not cloud-synced—so users can inspect what the agent has recorded. The settings panel shows a summary of captured activities, lets you delete individual entries, clear recent history entirely, or view the raw memory files. This design prevents the agent from accessing sensitive apps while maintaining full transparency about what it does see.

Automation and Personalization

Once the agent understands your workflow patterns, it can suggest automations based on repetitive work. In the demo, Computer History spotted that the user spent the morning drafting and refining scripts, prototyping prototypes, and triaging feedback—then summarized it accurately. The system can now generate skills to run automations like drafting a standup update every morning or tailoring tool use to match your habits. This personalization transforms ChatGPT from a question-answering tool into a workflow assistant that adapts to how you actually work rather than forcing you to adapt to it.

Takeaways

  • Enable Computer History in ChatGPT desktop settings on macOS to let the agent understand your actual workflow without manual context.
  • Review and manage captured memories in settings regularly—delete sensitive entries and blocklist apps you don't want the agent to track.
  • Ask the agent to suggest automations after it observes your repetitive patterns, then create skills to handle routine tasks like status updates.

Key moments

0:24Agent recalls last Google Doc

The last Google Doc you were editing was VideoScript Computer History earlier today.

0:47Codex contextualizes from earlier DM

Codex knows what conversation I was looking at earlier and can contextualize the message.

1:41Summarizes morning activity from events

This morning you mainly drove the Computer History launch video forward. Drafting, refining the script and iterating with teammates.

You just read one. Brevyd does this for every upload.

Follow OpenAI and every new video comes back as a summary like this, in your morning briefing. No watching required.