ChatGPT’s new Mac memory watches what you do, without screenshots
OpenAI on August 13 began rolling out Computer History, an opt-in feature for ChatGPT Work and Codex on macOS that turns your activity across apps and websites into memories and a timeline. The bigger change is how it gets that activity. As The New Stack reports from OpenAI’s announcement, Computer History records interaction events like clicks, typing and app switches. Chronicle, the experiment that came before it, took screenshots.
The goal is to stop re-explaining context. OpenAI’s own examples are “What was I debugging yesterday?” and “Where did I leave off on that PR?” ChatGPT uses the history to pull up the relevant files, documents and conversations, whether they sat in Google Docs, Slack or an open browser tab. The feature is for both products, which already share projects, chats, skills and plugins in one synced state. It is rolling out to Pro, Business and Enterprise users worldwide, except the EEA, the UK and Switzerland, where access comes “in the coming weeks.”
A Chronicle update, or a new product?
Testing Catalog framed the launch as “ChatGPT Chronicle gets an update with a new Computer History feature.” The New Stack treats Computer History as a separate feature. Its report describes Chronicle as a research experiment with the same goal that built its timeline from screen captures and OCR. It also separates the feature from Record & Replay, which lets ChatGPT and Codex users demonstrate a workflow and turn it into a reusable skill. Neither account disputes the goal. Where they differ is whether a new capture layer makes it a new feature.
The capture layer is the part that changed. Chronicle shipped on April 20 as a research preview inside the Codex macOS app, for Pro subscribers only, and needed the Screen Recording and Accessibility permissions. Sandboxed background agents saved PNGs to $TMPDIR/chronicle/screen_recording/ and deleted them after six hours. Selected frames went to a short-lived Codex session on OpenAI’s servers, which extracted summaries, OCR text, file paths and timing data. The results were stored as unencrypted Markdown in ~/.codex/memories_extensions/chronicle/ and injected into the system prompt of new Codex sessions. At the time, OpenAI warned that Chronicle could burn through rate limits quickly and raise exposure to prompt injection from whatever was on screen.
Computer History drops the screen entirely. “Computer History doesn’t rely on screen or audio capture,” OpenAI’s Dominik Kundel says in the announcement video. “Instead, it captures interaction events like clicking, typing, app switches, and more, so it can capture memories faster and more efficiently.” The New Stack’s view is that OpenAI “clearly learned from Microsoft’s Recall disaster.”
The retrieval side looks familiar. With Chronicle, OpenAI said that “when another source is better for the job, such as reading the specific file, Slack thread, Google Doc, dashboard, or pull request, Codex uses Chronicle to identify the source and then use that source directly,” per Help Net Security. Finding a file wherever it lives, as Computer History promises, appears to follow the same pattern: the memory points to the source rather than standing in for it.
What you control
Computer History is off by default. A new macOS menu bar item lets you exclude any app and pause capture, and a setting in the ChatGPT desktop app excludes apps and websites. The captured data is stored locally, so you can see what the system recorded and delete anything you don’t want on the timeline.
Raw events stay on the Mac, but the model still gets the summaries made from them. That is the catch The New Stack points to: it expects many users, and their employers, to leave the feature off for that reason, even without screenshots. Rollout scope has also widened. Chronicle launched for Pro alone and skipped the EU, the UK and Switzerland. Computer History arrives on all three paid tiers, and those regions only wait weeks.
Kundel pitches the feature as input for agents rather than a search box: “Once your agents understand what you do and how you work, it changes how you can engage with them. You can ask it to create skills based on what you just did, run an automation every morning to draft a standup update for your team, or simply tailor its use of tools to how you work.”
Up next
Claude Code 2.1.232: forked subagents are on by default
Claude Code 2.1.232 turns on forked subagents by default: a fork inherits the entire conversation and the prompt cache instead of starting cold. Type @ in the input line to message another of your sessions by name, routed through SendMessage.
Related stories
- OpenAI ships an official CLI for its REST API
- OpenAI sends GPT-4o users to Luna ahead of 2027 retirement
- OpenAI’s Decisions API beats its own engine by up to 10x
- OpenAI’s Decisions API picks an answer in 150 milliseconds
- HTTPX2 becomes the default in OpenAI’s Python SDK
- Claude Code on desktop folds an iPhone Duo to test layouts
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

