What ChatGPT Computer History actually is

Computer History is an opt-in feature in the ChatGPT desktop app on macOS. OpenAI shipped it on August 13, 2026. It turns activity in allowed apps and websites into a timeline and memories that ChatGPT and Codex can use.

It is not a screen recorder. It is not on by default. It is not on Plus, Go, or Free. It is not an API or Amazon Bedrock feature.

The official job is recent work context: pick up after a break, find a document you already had open, summarize the day, and notice a repeatable workflow that could become a skill or automation.

Who can turn it on

The docs list ChatGPT Pro, Business, and Enterprise. Each person opts in. Business and Enterprise first need an administrator to grant access. Admin grant is not the same as turning the feature on.

It also requires Memories. Use /memories to control whether a given chat can use local memories or write new ones. If Memories is off, Computer History does not run.

On August 20, 2026, the changelog added the European Economic Area, Switzerland, and the United Kingdom. The August 13 launch had excluded those regions.

What it records

Computer History builds an interaction-event stream from apps and websites you allow. Events can include clicks, typing, keyboard shortcuts, app switches, and text or other context that macOS exposes through accessibility.

It does not put screenshots in your history. It does not record microphone input or system audio. Private-mode browsing is never included. It does not need Screen Recording permission.

OpenAI says this replaces the earlier Chronicle research preview. It is a rebuilt system, not a rename.

The timeline lives under Settings > Computer history > History. Items group by day and time. Each can show a title, a text summary, the apps that contributed, and sometimes a suggested skill or automation. You can reveal the memory file in Finder or delete the item.

Example prompts in the docs:

• “What was I working on before my last break?”
• “Where can I find the proposal document I was looking for earlier today?”
• “Give me a list of tasks I’ve worked on today and their status.”
• “Prepare a summary of what I did yesterday for standup.”

The history is a pointer, not a dump of every file. When a Slack thread or a Google Doc is the better source, ChatGPT and Codex can use the timeline to find it, then read that source directly.

Controls

Under Settings > Computer history > Permissions you can exclude apps and website URLs, or allow only the ones you name. Clicking an app icon on a timeline item can exclude that app from future history.

Changing permissions affects future collection. Existing items stay until you delete or clear them.

The macOS menu bar ChatGPT icon can pause and resume collection, and can clear the last session for a recent app. In History you can clear the last 10 minutes, last hour, last day, or all history. Clearing deletes the matching events and the memories made from them. That cannot be undone.

OpenAI’s own warning: turn it off during communications with other people unless you have their prior express consent. Pause or exclude apps that hold health, financial, or personal information.

Where the files live

Temporary event files sit in the ChatGPT App Group on your Mac. Other apps cannot read them without permission. ChatGPT and Codex delete those event files after 48 hours.

The summaries become the same kind of local Codex memories: plain-text Markdown. The path is $CODEX_HOME/memories/extensions/skysight/, which is usually ~/.codex/memories/extensions/skysight/. You can read and edit those files. They stay until you delete them.

What OpenAI sees

Collection is local. OpenAI then processes the temporary event files on its servers so ChatGPT and Codex can write the memories. The docs say OpenAI does not keep those event files after processing unless required by law, and does not use them for training.

Later chats are a different bucket. If ChatGPT or Codex uses a memory, relevant memory contents and interaction events may go into that chat as context. That chat content may be used to improve models if your ChatGPT data controls allow it. Memories also follow the same per-chat controls as other Codex memories.

Summarizing activity uses tokens.

Prompt injection

The docs are explicit. Computer History raises prompt-injection risk from content in apps and websites. A page with malicious instructions can try to get ChatGPT or Codex to follow them. Treat untrusted pages the same way you would if the model could read the tab.

A skill or automation suggestion is a proposal. Review it before you ask Codex to create it.

What it is not

It is not Windows or Linux Computer History. The docs describe the macOS ChatGPT desktop app.

It is not a long-term notes app. The product language is recent activity: resume work, find a source, reuse a workflow. If you want a permanent knowledge base, that is still a separate store you maintain.

It is not on unless you turn it on, pick sources, and keep Memories enabled. Start with a short allowlist. Pause it for chat and banking. Read a memory file once so you know what “local Markdown” actually contains.

Source cutoff: Computer History and Memories docs, plus the August 13 and August 20, 2026 changelog entries, as retrieved August 22, 2026.