If you've noticed Claude or ChatGPT's desktop app using more system resources than you'd expect from what feels like a chat window, that's not an unreasonable observation; both companies have documented reports about exactly this.
What's been reported about Claude's desktop app
A GitHub issue on Anthropic's own repository describes disk writes of 8-13GB per hour from the Claude desktop app, alongside a separate reported issue of the renderer process using 80% or more CPU even while sitting idle without any active conversation. Sustained disk writes at that rate can contribute to a Mac feeling generally slower, since disk I/O and CPU load both compete with everything else your system is trying to do, even when the direct memory number itself might look moderate on the surface.
What's been reported about ChatGPT's desktop app
OpenAI's own community forum has a thread specifically titled around a memory leak in the Mac OpenAI ChatGPT application, and coverage from tech outlets has separately reported on ChatGPT's Windows client using unusually high memory with no apparent way for users to limit or control it. The Mac-specific reports are less extensively documented than the Windows coverage, but the underlying pattern, a desktop wrapper around a chat interface consuming more resources than users expect, appears on both platforms.
Why AI desktop apps tend to behave this way
Both apps are relatively new products built on similar web-technology foundations to other modern desktop apps, wrapping a chat interface, conversation history, and in some cases local processing or caching of context, inside a native app shell. Maintaining conversation history in memory for quick access, especially across long conversations or many separate chat threads, adds up in a similar way to how a messaging app's chat history accumulates. Because these are newer products than a mature browser or established productivity app, the kind of resource-optimisation work that's happened over years for other categories of software is comparatively recent here, which is consistent with both companies having open, acknowledged issues rather than a fully mature resource profile.
What genuinely helps right now
Keep the app updated. Because these are actively developed products responding to real user reports, checking for updates regularly is more likely to matter here than for more mature software, since resource-management fixes are exactly the kind of thing companies iterate on quickly once an issue is flagged publicly.
Restart the app periodically, especially after long conversations. If you've had one very long conversation thread open for hours, closing and reopening the app clears whatever context and history state has accumulated in memory.
Consider the web version for lighter everyday use. Both Claude and ChatGPT are available through a normal browser tab (claude.ai and chatgpt.com), which avoids the dedicated desktop app's specific resource profile, at the cost of losing any desktop-specific integrations like global keyboard shortcuts.
Report specifics if you're seeing this consistently. Both companies maintain public feedback channels (Anthropic's GitHub issues, OpenAI's community forum), and reporting your specific numbers adds to the evidence that helps these issues get prioritised for a fix.
Where a memory-relief tool fits in
Neither app's own resource use is something a general memory tool can directly reduce while the app is actively running; that requires the companies themselves to ship fixes. What Shiny helps with is everything else on your Mac competing for memory alongside an AI assistant app running in the background, particularly useful if you keep one of these apps open all day the way many people now do alongside their regular work tools.
Why this category is likely to improve faster than most
It's worth some optimism here that's specific to this app category. Both Anthropic and OpenAI are large, well-resourced companies actively competing for daily-use desktop presence, and public, visible resource complaints (GitHub issues, community forum threads, tech press coverage) are exactly the kind of feedback that gets attention quickly when a product is this actively iterated on and this publicly watched. Compare this to smaller, slower-moving software categories where a known issue can persist for years without action; the competitive pressure between major AI companies to have the smoothest, most trusted desktop experience gives them a strong incentive to address resource complaints faster than a niche utility might. That doesn't mean today's numbers will be fixed tomorrow, but the trajectory for this category looks more promising than for software without that same competitive and reputational pressure.
A sensible interim habit if you use both apps
If you regularly switch between Claude and ChatGPT's desktop apps, or keep both open for different tasks, treating them the way you'd treat any other pair of resource-hungry background apps is the practical approach for now: keep only the one you're actively using open, close the other when you switch tasks rather than leaving both running indefinitely, and don't assume either one is lightweight just because the interface itself looks simple. A chat window can feel deceptively minimal while the underlying app is doing meaningfully more work in the background than the visible interface suggests.