Historical Computer: When work context becomes a queryable memory
Listen to this article
Read by Anchor
What changes when the intelligent assistant does not start from the question you just typed, but from the trace of the work that preceded the question? That is the bet posed by the “Historical Computer” feature in the ChatGPT app for Mac desktop. According to coverage by Unite AI, the feature, once the user opts to enable it, records what they do across apps and websites, then turns it into memories and a searchable timeline, and ChatGPT and Claude can draw on them.
The idea does not mean that the conversation has become automatically smarter. Its practical meaning is that context may shift from the chat window to the work sequence itself: what was opened, what was repeated, and which path the user returned to. This represents a significant shift in how the assistant is envisioned, because the question is not necessarily the sole starting point. It also makes the boundaries of what the system collects and how the user interacts with it part of the experience, rather than a technical detail that can be postponed.
The memory here is a file owned by the user:The coverage says that the memories are saved on the user’s device in ordinary Markdown text files inside the Claude “memories” folder. The user can read and edit them directly. This gives the user a clear way to review what the tool retains, rather than dealing with memory as an opaque layer that cannot be accessed. However, it does not eliminate the need for the user to review themselves what they want to keep before building a later workflow on it.
The announced limits are no less important than the function:The feature does not capture screenshots, screen recordings, microphone input, or system audio, nor does it include browsing in private mode. These details directly affect the type of context that can reach the memory and the type of context that remains outside it. The available offer does not confirm the feature’s availability in Arab markets nor set a date for its release beyond the scope mentioned in the OpenAI post.
From recurring effect to skill:If the system detects a recurring action in the activity stream, one of the timeline points may suggest turning it into a reusable skill or a scheduled automation, and Claude then builds that from the recorded workflow. This is where the most sensitive value of the feature appears: not in merely saving the activity, but in suggesting that the repetition the user sees every day become an action that can be formulated and reused. For teams, the practical question is not whether the memory is convenient, but which recurring tasks deserve to become a skill, and who reviews that memory before it is adopted.
According to an OpenAI post, Historical Computer is available to Pro, Business and Enterprise subscribers in the European Economic Area, the United Kingdom and Switzerland. The details in this brief about registration, storage and automated skill suggestions are based on third-party coverage, so they should be treated as a description of the feature rather than a promise of its availability in every market.