OpenAI Computer History is designed to give ChatGPT and Codex a memory of what a user has been doing across apps and websites. Instead of relying on screenshots, the feature records interaction events and turns them into a searchable timeline that can later be used as context.
The idea is simple but significant: activity that would normally be scattered across windows, documents and browser sessions becomes organized memory. That can help ChatGPT find recent work, understand repeated routines and support Codex in turning those routines into reusable automation.
How Computer History Records Activity
Computer History replaces the earlier research preview called "Chronicle". The earlier version relied on screenshots, while Computer History works by recording events such as clicks, keystrokes, keyboard shortcuts and app switches.
OpenAI says the feature reads those events through macOS's accessibility system. The source material makes clear that screenshots, screen recordings, microphone input and system audio are not captured. Activity in private browsing mode is also never included.
The recorded events are periodically transformed into text summaries. Those summaries are stored locally as Markdown memory files, giving the system a written record rather than a visual archive of the screen.
The timeline view organizes these memories by day and time. It also shows which apps were involved, making it possible to search through recent activity in a more structured way than manually retracing steps across several tools.
Why The Timeline Matters For ChatGPT And Codex
The main value of OpenAI Computer History is that it gives ChatGPT and Codex context from real user workflows. According to OpenAI, typical uses include finding recently edited documents and creating standup summaries.
Because the activity has been summarized into memories, ChatGPT can use it to identify the right source and access it directly. That changes the role of memory from a passive note into a practical navigation layer across recent work.
The feature also looks for repeated workflows. When ChatGPT detects a pattern that appears reusable, the timeline can suggest a "Skill" or automation. A user can then ask Codex to build a reusable template from the recorded workflow.
That turns everyday repetition into something more actionable. Instead of only remembering that a task happened, Computer History can help identify that the same task happens often enough to be packaged into a repeatable process.
Opt-In Controls And Availability Limits
OpenAI built Computer History with several permission steps. In Business and Enterprise workspaces, an admin must explicitly enable the feature before anyone can use it. Even then, the feature does not turn on automatically for individual users.
Each user still has to consent on their own. The Memories feature also needs to be enabled, so Computer History depends on both workspace-level permission and user-level participation.
Users can decide which apps and websites contribute data by using include and exclude lists. Recording can also be paused or resumed at any time from the macOS menu bar.
The feature is not available yet in the European Economic Area, Switzerland, or the United Kingdom. The source article does not provide a timeline for availability in those regions.
Local Files, Plaintext Storage And Training Controls
OpenAI says temporary event files stay on the Mac inside the ChatGPT app group and are deleted after 48 hours. The Markdown memory files generated from those events stay on the local file system until the user deletes them manually.
Those memory files are not encrypted. That means another program running under the same macOS user account could potentially read them.
To generate the memories, OpenAI processes the temporary event files on its servers. OpenAI says it does not keep those files after processing, unless required by law.
OpenAI also says the files are not used for AI training. However, when memories are later included as context in chats, the contents of those chats may still become training data depending on the user's data control settings.
Prompt Injection And Consent Risks
OpenAI's own documentation flags a heightened prompt injection risk for Computer History. Because the feature captures content from apps and websites, malicious instructions embedded on a webpage could trick ChatGPT or Codex into following them.
That warning matters because Computer History is not only a memory tool. It can feed context into systems that take action or help build automation, so captured content may influence later behavior.
OpenAI also warns against using the feature with communication apps unless everyone in the conversation has given explicit consent. For apps that handle sensitive health, financial, or personal data, OpenAI recommends pausing recording or excluding those apps entirely.
The result is a feature with clear productivity potential and equally clear boundaries. Computer History can make ChatGPT memory more useful, but its value depends on careful app selection, explicit consent and an understanding that local Markdown memories remain readable on the same macOS user account.