OpenAI has introduced a new feature for ChatGPT's macOS desktop app called Computer History. The feature is designed to track clicks, keystrokes, and other actions, turning them into a timeline that ChatGPT and Codex can reference when you make a request. It is reminiscent of Microsoft’s controversial Windows Recall, but OpenAI is taking a different approach by relying on “events” rather than screenshots.
The feature is opt-in, not opt-out, meaning users must actively enable it before ChatGPT begins logging activity. Once enabled, it learns how you work, suggests automations, and can even pick up tasks you left half-finished. In a demo video, Dominik Kundel, a member of the Developer Experiences team at OpenAI, showed the app looking up the last document he edited, checking whether it had been shared with people via Slack, and delivering a recap of how he spent his morning.
What Is Computer History?
Computer History is a new layer of context for ChatGPT’s desktop app on macOS. Rather than simply responding to typed prompts, ChatGPT can now draw on a continuous stream of your computer activity. This includes which applications you use, which documents you edit, and which websites you visit — all organized into a searchable timeline. The system then uses that timeline to answer questions, suggest workflows, or automatically resume tasks that you may have abandoned mid-way through.
According to Ari Weinstein, Product and Engineering Manager at OpenAI, the feature will automatically ignore content in incognito or private browser tabs. Users can also exclude specific apps and websites from being tracked, and individual timeline entries can be deleted for finer-grained control. This is a notable departure from the always-on, default-enabled approach that got Microsoft’s Windows Recall into trouble.
How It Compares to Windows Recall
Windows Recall, introduced by Microsoft in 2024, was designed to create a comprehensive, searchable history of everything you do on a PC by taking screenshots every few seconds. The feature immediately sparked privacy concerns. Security researchers found that the database containing these screenshots was not encrypted on some early builds, and critics argued that silently capturing sensitive information like passwords, banking details, and private messages was dangerous. Microsoft delayed the rollout, redesigned some aspects, and eventually made it opt-in after intense backlash.
OpenAI’s Computer History is often described as “Windows Recall-like” because it aims to understand your digital life. However, there is a key distinction: Computer History does not capture images, videos, or audio. Instead, it relies on “events.” These events are likely structured records of user interactions — for example, “opened document X,” “typed in Slack,” or “visited website Y.” By avoiding raw screenshots, OpenAI sidesteps some of the most invasive privacy concerns associated with Recall. No visual recordings of your screen are stored, meaning passwords and other on-screen text are not captured as images.
Still, the feature remains “kind of creepy,” as the original description put it. Monitoring every click and keystroke is a significant level of surveillance, even if the data is used to improve your own productivity. The fact that it is opt-in and provides exclusion controls helps, but it also raises questions about where that data goes and how it is used.
Privacy and User Controls
OpenAI has emphasized that Computer History is opt-in, meaning you have to deliberately switch it on. That is a major difference from some other AI-powered productivity tools that enable monitoring by default. In addition to being opt-in, Computer History offers multiple layers of control. Users can exclude specific applications and websites from being tracked, ensuring that sensitive or private apps — like banking portals or medical record systems — never enter the timeline. Users can also delete individual entries, giving them the ability to scrub a particular action from history if they do not want ChatGPT to remember it.
Incognito and private browser tabs are automatically ignored, which is another safeguard. Private browsing is designed to leave no trace, and OpenAI has decided not to try to circumvent that. This is important because people often use incognito mode to search for health conditions, buy gifts, or do other things they do not want saved into a persistent profile.
However, there are still unresolved questions. OpenAI has not shared detailed information about where Computer History data is stored, whether it is encrypted, or how long it is retained. It is also unclear whether the data is used exclusively for your own personal context or if it is also used to train OpenAI’s models. The original announcement mentioned “turns your actions into training data,” which suggests that the activity timeline may be included in the broader AI training datasets. That would be a significant privacy consideration.
How ChatGPT and Codex Use the Timeline
Computer History is not just a passive log. It is designed to actively assist you. For example, in the demo video, Dominik Kundel asked ChatGPT to look up the last document he edited. The system was able to access the timeline, identify the most recent document, and then check whether it had been shared via Slack. That kind of cross-application awareness is a major step forward for AI assistants, which typically operate in isolation without access to your broader digital environment.
Codex, OpenAI’s coding tool, can also reference the timeline. If you are a developer working on a project, Codex could look at your recent coding sessions, see which files you were editing, and help you pick up where you left off. This could be particularly valuable for complex tasks that involve multiple steps and tools, such as writing a script, running tests, and then documenting the results.
Automations are another key use case. By learning your habits, ChatGPT can suggest shortcuts or even automate repetitive workflows. If you always open the same set of applications at the start of the day, or if you frequently copy data from one app to another, ChatGPT might offer to do that for you. This moves ChatGPT beyond a simple chatbot and into the realm of an agentic assistant, one that can anticipate your needs and act on your behalf.
The Broader Context of AI Assistants
Computer History is part of a broader trend toward AI assistants that are more deeply integrated into our computing environments. Both OpenAI and Microsoft are racing to create “agents” that can perform tasks across multiple applications. OpenAI has been pushing ChatGPT with features like Advanced Voice Mode, while Microsoft has been embedding AI into Windows and Office through Copilot. The idea is that AI should not just answer questions but also take action, and to take action it needs context about what you are doing.
But context is a double-edged sword. The more an AI knows about your activities, the more useful it can be — but also the more it intrudes on your privacy. Windows Recall showed that the market is wary of AI-powered surveillance, even when the intention is helpfulness. OpenAI appears to be trying a more careful approach, but the underlying mechanism is still essentially the same: the AI watches everything you do.
The response to Computer History will likely hinge on trust. OpenAI has faced privacy scrutiny in the past, including concerns about ChatGPT’s ability to remember conversations and the potential for data leaks. With Computer History, the stakes are higher because the AI is not just observing your chats; it is observing your interactions with every part of your computer.
For now, Computer History is a bold experiment in turning personal activity into a resource for AI. It offers tangible benefits, such as automated task resumption and intelligent recaps. But it also forces users to make a choice: how much of your digital life are you willing to log in exchange for that convenience? With careful controls and transparency, OpenAI could set a new standard for privacy-conscious AI assistance. Yet given the creepy factor, the company will need to earn user trust one click at a time.
Source:The Verge News