StoweMost AI planning apps are cloud apps: the text or audio you give them travels to a server, a model there does the thinking, and the result comes back. A smaller group runs the AI on the phone itself, so nothing you say or type leaves the device. Both architectures produce a to-do list; they have very different answers to the question "who else could ever read mine?" — and apps are rarely eager to tell you which one you are using.
This post is the check I wish existed when I was researching the category: what each architecture actually means, and how to tell them apart from the outside in five minutes.
What happens to your tasks in a cloud AI planner?
Your words are transmitted, processed, and — for some window, under some policy — stored somewhere you cannot see. That is not an accusation; it is what "AI in the cloud" means. Concretely, the trip usually looks like: your text or audio goes to the vendor's server, often onward to a third-party model API the vendor rents, gets processed, and comes back. Along the way it may be logged, cached, retained for "service improvement", and covered by more than one company's privacy policy.
A well-run vendor minimizes all of this, and many do. The point is structural: in a cloud architecture, privacy is a policy — a promise that transmission, logging, and retention are handled well. Policies can be good, and they can change, be breached, or be acquired along with the company. You are trusting conduct you cannot observe.
What does on-device AI change?
It changes privacy from a policy into a property. When transcription and the model both run on the phone's own chip, your words are never transmitted, so there is nothing to log, nothing to retain, nothing to breach at a server, and no policy update that can quietly widen what happens to them. The vendor cannot mishandle what the vendor never receives.
The trade-offs are real and worth stating plainly. On-device models are smaller and less capable than frontier cloud models — fine for structuring what you said into tasks and dates, not for open-ended reasoning over your whole history. They need recent hardware; on iPhone, roughly the Apple Intelligence line and newer (I keep a reference table of which iPhones run local AI). And sync must go through something like the platform's own encrypted cloud rather than a vendor server, which usually means no web app.
How do you tell which architecture an app uses?
Five checks, cheapest first:
- The airplane-mode test. Turn off Wi-Fi and cellular, then use the AI feature. If it fully works, the processing is on-device — there is nowhere else for it to happen. This is the one test that cannot be worded around.
- The account test. An app that requires sign-in before working has a server relationship at its core. No-account apps have far fewer places to send your data.
- The App Store privacy label. "Data Not Collected" or "Data Not Linked to You" are the strong tiers. "Data Linked to You" on a planner deserves a pause.
- The privacy policy word search. Search for "third part", "model provider", "retention", and "service providers". A policy that names a rented model API is telling you your tasks make two hops, not one.
- The pricing sniff test. Servers and rented model calls cost money per user. A free AI feature with no visible business model is being paid for somehow; it is fair to ask how.
When is cloud the right choice anyway?
When the capability is the point. If what you want is an assistant that reasons over months of context, drafts replies, or juggles a team calendar, cloud models are simply better at it, and vendors like Motion or Reclaim are honest about being cloud services — that is what makes their features possible. Sync through a vendor server is also what gets you a web app and Android-plus-iPhone in one product, which platform-bound on-device apps mostly cannot offer.
The dishonest position is not "cloud"; it is "cloud, vaguely". If a vendor is proud of the architecture, the policy says so plainly. The pattern to walk away from is AI marketed as magic with the data path left as a footnote.
Where does Stowe sit in this?
Stowe is my attempt at the on-device end of this trade, so judge the description accordingly: transcription and the sorting model run on the iPhone (Apple Intelligence), every feature works in airplane mode, there is no account and no server of mine, and data lives on the device and in your private iCloud — the App Store label reads "Data Not Linked to You". The costs I listed above apply in full: it needs a recent iPhone, and there is no web app. I wrote up the mechanics in an honest look at what offline means. If you take one thing from this page, take the test rather than the recommendation: airplane mode on, mic button pressed. Whatever happens next is the app telling you the truth about itself.
Related: What offline actually means in an AI planner · Which iPhones run local AI · Voice to-do list apps in 2026
Frequently asked questions
Does "we don't train on your data" mean an app is private?
No — it is a narrower promise than it sounds. Not training on your data still allows sending it to a server, processing it there, logging it, and retaining it under a policy you have to trust. It rules out one use, not the transmission. The stronger claim is architectural: the data never leaves the device. That one is checkable — airplane mode.
How can I check whether an app processes my data on-device?
Airplane mode is the honest test. Turn off Wi-Fi and cellular, then use the AI feature. If it works fully, the processing is happening on the phone — there is nowhere else for it to happen. Also check the App Store privacy label ("Data Not Collected" or "Data Not Linked to You" are the strong tiers) and whether the app works without an account.
Is cloud AI in a planner always a bad thing?
No, and it would be dishonest to say so. Cloud models are larger and more capable than anything a phone runs, and features like cross-device sync through a vendor server or a web app genuinely require servers. The question is not cloud-bad, it is whether you know your task list is making the trip, and whether that trade was worth it for what you got back.
What is actually sensitive about a to-do list?
A real task list is a diary that doesn't know it. Health appointments, medications, money worries, job interviews while employed, relationship repair, the names of your children and doctors — spoken in your own words, timestamped, updated daily. It deserves the same caution as messages, and it rarely gets it.