OpenAI has switched on ChatGPT Computer History, an opt-in feature in the ChatGPT desktop app on macOS that records what you do across your apps and websites and turns it into memories and a searchable timeline ChatGPT and Codex can draw on. It was announced in the desktop app changelog on 13 August 2026, and it is available to ChatGPT Pro, Business and Enterprise users. India is included from day one. The European Economic Area, Switzerland and the United Kingdom are not.
The feature replaces Chronicle, OpenAI’s earlier research preview, but the company describes it as a rebuilt system rather than a rename. Chronicle worked off screenshots. Computer History does not take screenshots at all — it reads interaction events that macOS exposes through its accessibility system: clicks, typing, keyboard shortcuts and app switches. It periodically bundles those events into text summaries and local memory files.
The pitch is straightforward enough. You can ask what you were working on before lunch, find a document you half-remember opening, or get a summary of yesterday for standup. When it spots repeated work, a timeline entry can suggest turning it into a skill or an automation.
Ari Weinstein, whose X bio describes him as working on “Computer Use & Frontier Interfaces @OpenAI”, put it bluntly when announcing it:
Today we’re releasing Computer History in ChatGPT. It lets ChatGPT learn from everything you do on your computer, so it can better understand how you work, finish tasks that you’re in the middle of, and suggest skills and automations based on how you use your computer.
What ChatGPT Computer History actually sends to OpenAI
The framing around this feature leans hard on the word “local”, and that word is doing more work than it should. The memory files are local — plain-text Markdown under ~/.codex/memories/extensions/skysight/, which you can open and edit yourself. The raw interaction events are stored on your Mac too, inside ChatGPT’s App Group container, and deleted after 48 hours.
But the summarisation does not happen on your Mac. OpenAI’s own documentation says the company “processes temporary event files on its servers to generate memories, which are then stored locally on your Mac”. So a stream describing your clicks, typing and app switches does leave your machine. OpenAI says it does not retain those event files after processing, unless required by law, and does not use them for training.
That last assurance is narrower than it sounds, and it is worth reading twice. It covers the event files. Separately, the documentation notes that when ChatGPT or Codex uses a memory in a later chat, the memory contents and interaction events may be included as context in that chat — and that chat content may be used to improve OpenAI’s models, subject to your ChatGPT data controls. “Not used for training” applies to one part of the pipeline, not the whole of it.
OpenAI’s own warnings are the ones to read
To OpenAI’s credit, the documentation is unusually candid about the risks, and the warnings there are more useful than any outside critique.
- The memory files are not encrypted. OpenAI’s own alert states that Computer History files “can contain sensitive information”, are not encrypted by the feature, and that “other programs running as your macOS user may be able to access them”. Any app you run under your own account can potentially read a plain-text log of your working day.
- Other people have not consented. The documentation tells users to turn the feature off “during communications with other people unless you have their prior express consent”. That puts the entire consent burden on you, and gives the person on the other end of your Slack thread or work chat no notice and no control — the same asymmetry we noted in Twitch’s default-on AI training change earlier this week.
- Prompt injection gets easier. OpenAI states plainly that Computer History increases the risk of prompt injection from content in apps and websites — visit a page carrying malicious instructions and ChatGPT or Codex might follow them.
- Private browsing is excluded, and so are screenshots, screen recordings, microphone input and system audio.
The regional split is telling on its own. India, the United States and most other markets get this immediately; the EEA, Switzerland and the UK — the jurisdictions with the most demanding data-protection regimes — do not. OpenAI gives no reason for that, and this is our reading rather than the company’s, but the pattern of shipping a continuous-activity recorder everywhere except the strictest privacy jurisdictions first is hard to miss. Indian users get the feature and the compliance questions at the same time.
How to turn it on, pause it, or clear it
Computer History is off by default for everyone, and administrator approval in a Business or Enterprise workspace does not opt anyone in — each person still has to enable it themselves. It also requires Memories to be switched on, and is not available with an API key or through Amazon Bedrock.
- Open the ChatGPT desktop app on macOS.
- In Settings, under Integrations, select Computer history.
- Select Turn on and review the privacy, permissions and local-storage information.
- Turn on Memories if prompted.
- Choose which apps and websites can contribute, then follow the macOS permission prompts. Screen Recording permission is not required.
Under Settings > Computer history > Permissions you can either exclude specific apps and websites or flip it around and include only the ones you choose — the second option is the safer default if you keep banking, health or personal material on the same machine. Changing permissions only affects future history. To control collection in the moment, use the ChatGPT icon in the macOS menu bar, which has Pause and Resume controls and shows what is being captured.
To clean up, open Settings > Computer history > History. You can reveal any summary’s memory file in Finder, delete individual timeline items, or clear the last 10 minutes, the last hour, the last day, or everything. Clearing deletes both the interaction events and the memories built from them, and cannot be undone.
It is a genuinely useful idea, and OpenAI has built more controls into it than most features of this kind get. But a tool that logs your typing and clicks all day into unencrypted plain-text files, ships those events to a server for processing, and asks you to personally obtain consent from everyone you talk to, deserves a deliberate decision rather than a reflexive “Turn on”. If you enable it, spend the extra minute on the include-only list first.
Sources: OpenAI Computer History documentation, 9to5Mac
