ChatGPT Computer History: Hands-On Setup Guide and Privacy Risks
A hands-on look at ChatGPT's Computer History feature: how app inclusion settings work, what it actually captures, and the privacy trade-offs to weigh.
What is ChatGPT’s Computer History feature?
Computer History is a new setting inside the ChatGPT desktop app that lets ChatGPT track your activity across the apps and websites you use on your Mac, then summarize what you did, answer questions about your work, and suggest tasks worth automating. OpenAI says it does this without recording your screen or audio, relying instead on activity logs tied to the apps you choose to include. It’s currently limited to the Mac desktop app, requires a Pro, Business, or Enterprise plan, and is not available in the EU, Switzerland, or the UK.
TL;DR
- Computer History lets ChatGPT observe activity across apps and websites you approve, then summarize your day, answer follow-up questions, and suggest automations without recording video or audio.
- The rollout is narrow: Mac-only, paid-plan-only (Pro, Business, or Enterprise), and geo-restricted out of the EU, UK, and Switzerland at launch.
- Setup works through an allowlist or blocklist model in the ChatGPT desktop settings, where you choose to exclude specific apps or, more restrictively, include only specific apps.
- In hands-on testing, the feature correctly reconstructed a multi-app workflow (Google Docs, ChatGPT, Finder, Gmail) and generated a reusable “skill” from it, plus a list of tasks worth automating.
- The privacy trade-off is real: giving ChatGPT access to a browser like Chrome effectively exposes most of a person’s digital life, and the default settings require active pruning to avoid over-sharing.
- This feature closely resembles Microsoft’s Recall, which faced heavy backlash in 2024 for similar always-on activity tracking, suggesting the same scrutiny may follow ChatGPT’s version.
- The bigger implication is architectural: agentic AI has lacked a way to learn from real, sequential human workflows, and Computer History is one of the first mainstream attempts to close that gap.
Remy doesn't write the code. It manages the agents who do.
Remy runs the project. The specialists do the work. You work with the PM, not the implementers.
How does Computer History actually work?
Once enabled in Settings under the Integrations tab, ChatGPT begins logging activity in the apps you’ve approved. It’s not a screen recording. Instead, it appears to track things like which app is active, what content is visible or typed, and how tasks move across applications over time. In one test, the system correctly summarized a session that involved drafting a document in Google Docs, asking ChatGPT to reformat it into a project plan, saving that plan as a PDF through Finder, renaming and relocating the file, and sending it as an email attachment through Gmail in Chrome.
Asking “what have I done today?” produced an accurate rundown of that entire sequence. Asking ChatGPT to turn a finished workflow into a reusable “skill” worked as advertised: it generated a packaged process based on the actual steps taken, not a description typed in after the fact. That’s a meaningful difference from how ChatGPT has worked until now, where it only knew what a user explicitly told it. With Computer History, it can infer process directly from behavior.
The feature also supports a more open-ended prompt: asking which tasks from the day could be automated. In testing, it correctly flagged document-based work, PDF generation, and repetitive checks as “highly automatable,” and even referenced specific to-do items that had been typed inside a document during the session, not just the fact that a document was opened.
How do you set up app inclusion and exclusion?
Inside ChatGPT’s desktop settings, the Computer History tab offers two configuration modes. The default approach is exclusion-based: every app is tracked unless you manually add it to an exclude list. The alternative, described as more manageable in practice, flips this to inclusion-only: nothing is tracked except the specific apps you name.
The exclusion-first default is awkward because it puts the burden on the user to anticipate every sensitive app in advance, things like Messages, health apps, finance apps, or anything else that could expose private information. Missing one means it gets tracked by default. Switching to an inclusion-only list (in testing, this meant approving only Finder, TextEdit, and Chrome) is a more deliberate way to limit exposure, though it still means anything happening inside an included app, including a browser, is fair game.
OpenAI’s own onboarding language recommends pausing the feature during sensitive activity like video calls or messaging conversations with other people, and says users are expected to have “prior express consent” before capturing conversations involving anyone else. That’s a meaningful caveat: enabling this feature broadly could mean capturing details of other people’s communications without their knowledge unless it’s manually paused first.
Is Computer History worth the privacy trade-off?
Remy is new. The platform isn't.
Remy is the latest expression of years of platform work. Not a hastily wrapped LLM.
That depends heavily on what apps get included and how much of a person’s digital life runs through them. Including a full browser like Chrome is close to giving ChatGPT visibility into most day-to-day work: email, documents, social media, banking portals, search history, and more. For anyone who has not already drawn a personal line around what they’re comfortable sharing with an AI company, this feature forces that decision immediately.
The upside is significant. This is one of the first mainstream tools that gives an AI system context not just about what a person says they’re doing, but the actual sequence of actions taken across multiple applications over time. That sequential, real-world context is something most memory systems and AI agent frameworks have lacked entirely, since they typically only see what’s explicitly fed to them through prompts, connectors, or manually maintained memory stores.
The comparison that keeps coming up is Microsoft’s Recall, a Windows feature announced in 2024 that also aimed to capture ongoing activity for later recall, and which drew immediate, intense backlash over privacy concerns before Microsoft scaled back its rollout. Computer History arrives with some of the same design tension: powerful context-building capability paired with the risk of normalizing constant activity logging. Whether it faces the same backlash, or gets a pass because of how it’s scoped and gated behind paid plans and manual opt-in, remains to be seen.
Why does this matter for the future of AI agents?
AI agents are only as useful as the context they have. Techniques like skills (packaged prompts for repeatable tasks), connectors (pulling in data from other tools), and memory systems (recalling past interactions) all exist to solve the same underlying problem: agents don’t know what a person actually does unless it’s explicitly given to them.
Computer History approaches that problem differently, by observing real workflows directly instead of relying on descriptions after the fact. That matters because the hardest part of using agentic AI has never really been building an agent, it’s been figuring out what to hand off to it in the first place. A system that watches how someone actually works and then proposes what to automate turns that guesswork into something closer to an audit.
Paired with simpler, chat-based interfaces for interacting with AI agents (a trend also visible in other recent consumer AI products built around messaging-style agent interactions), this kind of activity-aware context could make agentic workflows far more approachable for people who aren’t already comfortable with terminals, APIs, or manually maintained memory systems.
Frequently Asked Questions
What platforms support ChatGPT’s Computer History feature?
It is currently available only in the ChatGPT desktop app on Mac, and requires a Pro, Business, or Enterprise subscription. It is not available to users in the EU, UK, or Switzerland.
Does Computer History record my screen or audio?
According to OpenAI, no. It tracks activity across approved apps and websites to build a summary of what was done, but does not capture screen recordings or audio.
Can I control which apps ChatGPT tracks?
Yes. Settings allow either an exclusion list, where all apps are tracked except the ones you name, or an inclusion-only list, where nothing is tracked except the apps you explicitly approve. The inclusion-only approach gives more precise control.
How is this different from Microsoft Recall?
Both features aim to give an AI system persistent awareness of user activity over time. Microsoft Recall faced significant public backlash over privacy concerns after its 2024 announcement. Computer History launches with narrower availability (Mac only, paid plans, opt-in) and stated recommendations to pause tracking during sensitive activities, but raises similar underlying questions about how much activity data should be captured by default.
Should I turn on Computer History?
That depends on individual privacy tolerance. Including broad tools like a web browser can expose a large share of daily digital activity to the system. Reviewing the inclusion/exclusion settings carefully, and considering which apps genuinely need to be tracked, is worth doing before enabling it.