OpenAI has released the Computer History tool for the desktop ChatGPT application on macOS. It analyzes user activity across different programs and websites to make the model's subsequent responses more accurate and personalized.
The update was released on August 13th and is available for Pro, Business, and Enterprise plan subscribers. Residents of the EEA, the UK, and Switzerland will gain access later — within the next few weeks, the company plans to bring the feature into compliance with local data processing requirements.
The company announced the launch on social network X, calling Computer History an evolution of the earlier experimental project Chronicle. According to OpenAI, the new version consumes fewer tokens to maintain context and offers users more privacy control tools than the early test release.
How It Works
Computer History is built into the ChatGPT desktop app for Mac and runs in the background. The service records which applications and websites a person uses and compiles this data into a temporary activity timeline. In subsequent dialogues, the model can rely on this history, reducing the amount of context that would otherwise need to be entered manually.
According to the company, the timeline interface allows users to return to previously completed tasks, track recurring work patterns, and clear the entire history or only selected periods — either through the timeline itself or via the menu in the macOS status bar.
The key condition is enabling the option in the Settings section, under the Integrations tab. Without this step, data collection does not begin, and the function's status in the settings is always displayed openly so the user can see whether tracking is active.
OpenAI describes the Computer History function not merely as an extension of memory about conversations within ChatGPT, but as a mechanism that allows the model to account for a user's recurring work patterns outside of dialogue with the chat-bot.
ChatGPT's Memory Expands, Europe Waits for Now
In 2026, OpenAI significantly accelerated the implementation of persistent memory across all its key products. The experimental Chronicle function, which is now being replaced and enhanced by Computer History, already introduced the idea of cross-session context for desktop version users.
The new release reduces the computational cost of maintaining such context. On the same day, OpenAI announced GPT-5.6 Ultrafast mode on Cerebras chips — for API clients, generation speed reaches 750 tokens per second. This was announced just a few hours before the launch of Computer History.
At the same time, regulators in the EU and the UK are closely monitoring tools that involve the permanent storage of user data. The decision to postpone the launch of Computer History in the EEA, the UK, and Switzerland indicates that the company first aims to bring the feature into compliance with local regulations before granting access in these regions.
Users who do not want the system to track their activity can simply not enable this function — it operates solely on the principle of voluntary consent.
AI Opinion
From the perspective of machine data analysis, the predecessor to the new feature contained technical risks not mentioned in the article. The official OpenAI documentation for the Chronicle project directly indicated an increased risk of prompt injections and the storage of records in unencrypted form on the user's device — these details are important when assessing the maturity of the technology underlying Computer History.
The macroeconomic context here also deserves attention. The trend towards constant personalization through the analysis of user activity intensifies competition with other major players in the chatbot market — in particular, with Anthropic's Claude and Google's Gemini, which are developing similar memory mechanisms in parallel with OpenAI. The delayed launch in the EEA, the UK, and Switzerland can be viewed not only as a gesture of compliance with regulations but also as an indicator of how difficult it is to balance deep personalization with data protection requirements across different jurisdictions.
Will voluntary consent remain a sufficient barrier for user trust as the volume of collected activity data grows?
end-content
