How it works, explained
4 min read · Checked September 17, 2026
Why accountability needs no screenshots
Screenshot accountability captures your screen, uploads it, and gives someone a feed to review. On Guard looks at your screen on the device, in memory, discards the frame, and sends one text that says it happened. No frame is saved or uploaded unless you choose to send us one screenshot after a false alarm, and that one never reaches your ally. We think that is the safer model and the more human one. Here is the reasoning.
On this page
Two models of accountability
Two ways to do accountability
Screenshot accountability
Your screen
Captured on a schedule.
Uploaded
Images leave the device.
A feed
Someone reviews your screens.
What travels: pictures of what you saw. What the other person carries: a review job.
Detection and check-in
Your screen
Classified on the device.
An event
App, device and time. Never the site.
A text
Your ally reaches out.
What travels: one short message. What your ally carries: a reason to call.
Both models want the same thing: a moment that would have been private is not. They get there differently. One moves pictures of your screen to a server and then to a person. The other moves a verdict, with no picture, to a person. The rest of this page is why we chose the second.
What a screenshot report exposes
- Everything else on your screen at that moment: a bank tab, a message thread, a name, a medical page. Blurring helps; it does not make the image not exist.
- A store of images that has to be secured forever, by someone other than you.
- A person who now carries a review job, and who learns things about you that were not the point.
- A gallery of your worst moments, kept. That is a strange thing to build for a man who is trying to move on.
What a check-in exposes
Three facts: a known adult site was opened, in which app, on which device and at what time. Your ally does not get the site, an image, a category, or a count. He gets a reason to reach out, which is the only thing he needed.
The promise
Allies are alerted with the app, the device and the time. No additional details.
It also keeps the relationship right. A brother who gets a text can be a brother. A brother who gets a feed becomes an auditor. Being your ally should cost him almost nothing, so that he says yes and stays.
The privacy boundary, in plain words
- Detection runs on the device and the frame is discarded after the verdict. No pixels leave the device unless you send one screenshot yourself.
- Our servers hold your account, your ally links, your settings, your subscription state, a liveness heartbeat, and payload-free detection events. Not URLs, not history, and no screenshot you did not send.
- If a gate is a false alarm, you can tap "Think this is a mistake?" and then, with a second, separate tap, send us that one screenshot. It is your whole screen at that moment, with the outline On Guard drew and the detection scores, linked to your account. Only On Guard admins can open it, every view in our review tool is logged, and it is deleted after 90 days. Your ally never gets it.
- The backend cannot unblock anything, and a billing problem never reaches an ally.
- The check-in text is composed from your display name only. There is no field for anything else.
What on-device detection means The exact list of what does and does not cross.
The privacy policy The same boundary, in the legal document.
The trade, stated honestly
The honest line
A screenshot feed can show a reviewer things a classifier misses, and it can be dressed up as proof. We give that up. What we keep is a system with no screenshot feed to leak, a text your ally can act on in ten seconds, and a man who is interrupted and supported instead of documented.
Blocking alone fails. Any man can get around a blocker, ours included. That is why the redirect and the ally exist. Neither of them needs a picture of your screen to work.
Questions
Does On Guard ever save a screenshot, even locally?
Not on its own. On Guard holds each frame in memory only as long as it takes to classify it, then discards it. Checking a frame never writes it to disk or uploads it. The one exception is a screenshot you choose to send us after a false alarm: a second, separate tap sends that one image, linked to your account, and we delete it after 90 days. On Android the setup sheet says screenshots are taken automatically and checked on the phone, in memory.
How can my ally trust the text without seeing proof?
The text is the proof that a gate fired. What it does not carry is the content, on purpose. Trust between you comes from the conversation, not from a picture.
What if I want my ally to see more?
That is not something On Guard can do, by design. There is no setting that sends more, and no path in the system for it.
Do you compare On Guard with screenshot-based apps anywhere?
Yes, on the comparison pages, with named products and checked facts. This page keeps to the mechanism.
Checked September 17, 2026 against the shipping app. If something here no longer matches what you see, email team@stayonguard.app.
Read next