ShotDetect is a research prototype that listens for the sound of gunfire. It is not an emergency service and it cannot bring help. It warns the people standing near it whose phones are running it too, and it can forward a corroborated alert to a guardian you have paired with, wherever they are. This paragraph used to end "nobody further away than Bluetooth reaches is told anything". That was true until August 2026. Clause 9 sets out what that forwarding does and does not promise. It is deliberately quieter than the alert shown at the school, and hearing nothing from it tells you nothing.
Since August 2026 it also shares your live position with the people you paired with, while the app is open, in both directions. That is a separate thing from an alert and it is set out in clause 1a. Version 1.2 of these terms described a position that travelled only inside a warning. That was true then.
By using the app you agree to what follows. If you do not agree, do not use it.
What this app does
ShotDetect listens for loud, sudden sounds through the microphone of the phone it is installed on, and scores each one with a trained classifier to judge whether it sounded like gunfire. When it decides that it did, it broadcasts that judgement — with the moment of the sound and where the phone was — to any phone within Bluetooth range, and it listens for the same from them. A single phone never raises an alarm. It takes at least three phones that heard the same sound at the same moment, from points at least fifteen metres apart; below that the screen says which condition was not met rather than warning you.
One thing in the design is not built and two are only partly proven. There is no path by which an alert reaches a school or an emergency service; that describes where the app is going, not where it is today.
Reaching a guardian who is not nearby was in that list until August 2026 and is now built. A corroborated alert is encrypted on your child's phone and forwarded through a relay we operate to the guardian you paired with, wherever they are. It is partly proven in the same sense the Bluetooth link is. An alert has been measured crossing between two real handsets — arriving in well under a second on a phone that was asleep with the app closed — but both phones were on the same desk, reaching a relay on a cable. How often it arrives, and how quickly, across the real internet to a parent in another country has not been measured. Nor has whether a phone keeps watching across a whole school day.
It is also deliberately quieter. A detection must clear a substantially higher confidence bar before it is forwarded to someone who is not at the school than before it is shown to someone who is. Measured against held-out audio, that reduces false alarms to roughly one per school per year, and means about two in every five of the detections that would raise an alert at the school are never forwarded at all. Silence is not evidence that nothing happened.
The second partly-proven thing is the Bluetooth link: a message has been shown to cross between two handsets on a desk, in under a second and in both directions, but how reliably it crosses a real building — through walls, along a corridor, with many phones at once — has not been measured. Do not install this expecting to be warned about anything.
Paired people see where each other are
This is new in August 2026 and it is the largest change to what this app knows about a child since it began reading a location at all. These terms, and clause 9 below, described a position that travelled only inside a warning and never between warnings. That was true when it was written.
What happens now: while the app is open on a phone, that phone shares where it is with the people it paired with, about every five seconds. Each update carries the position, how accurate it is, the time, and whether the phone is moving. On the other phone it draws a dot on a map. It is not a notification, it makes no sound, and no history of it is kept.
Who can see it. Only people you paired with, in person, by showing each other a code. A student's position goes to that student's guardians; a guardian's position goes to their students. There is no directory, no search, no proximity list, and no way for one family's phone to find another's. Pairing with several people on each side is normal — a child may be paired with two parents and a grandparent, and each of them sees the same thing.
It goes both ways. Alerts travel one way, from a student's phone to a guardian's. Positions do not. A guardian with the app open is showing their children where they are, on the same terms.
When it stops. Closing the app stops it — nothing is published while the app is not on the screen of the phone doing the publishing. The exception is an alert, which sends a position whether the app is open or not. A position nobody has updated for ninety seconds is removed from the map rather than left there looking current.
What we do not promise. That any of it arrives, that it is accurate, or that it is current. A phone with no signal, no location permission, a poor fix or a flat battery shows nothing or shows something vague. An empty map is not evidence that anyone is anywhere, and this app must not be relied on as a way of knowing where a child is. Clause 9 applies to a position exactly as it applies to an alert.
What it cannot do
It cannot call the police, an ambulance or a school. It cannot summon help of any kind. It shows you your local emergency number, and it will never dial it for you.
It will sometimes be wrong
This app may warn you when nothing has happened, and it may stay silent when something has. A phone in a bag, a flat battery, a denied microphone permission, a school with no other phones running it, or simply a sound the app does not recognise will all produce silence.
Treat every alert as unconfirmed until you know more.
Silence from this app is not evidence that nothing happened.
Do not rely on it alone
ShotDetect is not a substitute for any part of an emergency plan, and a school should not treat it as one. It is an extra signal, and an unreliable one.
Who may use it
Protection does not begin until a guardian has connected to the student's phone and accepted these terms, so a child never agrees to this alone.
That gate is enforced: a phone will not arm unless a guardian has accepted on it, so a student tapping through the disclaimer alone cannot open the microphone. This page said the opposite until 2026-08-18, describing a gap that had already been closed. What the app cannot do is check who is tapping, which is a different limitation and is stated on the screen that asks.
Use by a child under 13 requires verifiable parental consent.
Calling emergency services
If you call emergency services after an alert, say that an app alerted you and that it is not confirmed. That matters: the people answering need to know what kind of report they are receiving.
Making a false report to emergency services is a criminal offence in most countries, including the Philippines.
Practice alerts
The app can run practice alerts so people know what a real one looks like. A practice alert is marked as practice on every screen it appears on, and it cannot open the emergency dialler — there is no path from a practice alert to summoning the police.
A practice alert does appear in the app's timeline, labelled as a drill, and the entries stay there after the practice ends. The timeline is deliberately append-only, so that someone reading it afterwards sees the whole sequence rather than a record wiped the moment the screen cleared. Because the timeline lives in memory only, it is gone when the app closes.
The screenshot of an alert on our home page is a practice alert, captioned as one. The app cannot currently produce that screen from real audio.
No guarantee of detection
We do not guarantee that the app will detect a gunshot, that it will detect one in time, or that it will tell the difference between a gunshot and another loud sound. It is a prototype and its accuracy in real buildings has not been established.
What has been measured, on recorded audio rather than in a building, is set out on our early access page. In short: the detector would raise roughly 30 false alerts per school year against a design target of about one, and on a phone microphone it was never trained on it detects about 56% of gunshots — less still on other kinds of recording equipment it has never encountered. Neither figure is good enough for anyone to rely on. We publish them so that nobody has to discover them the hard way.
Delivery is not in our control
This clause used to begin "No warning reaches anyone out of Bluetooth range, today or in this version." That is no longer the case: since August 2026 a corroborated alert can be forwarded to a guardian you have paired with, wherever they are. What follows is what that does and does not promise.
All of it applies to the position sharing in clause 1a as well. A position that does not arrive, arrives late, or arrives stale is the same failure as an undelivered alert and is no more under our control. The app draws nothing where it has heard nothing, on purpose, but the absence of a dot tells you about the network and not about a person.
Between phones in Bluetooth range, whether a warning arrives depends on batteries, radio conditions, walls and distance, notification settings, and whether the phones involved are switched on, armed and nearby. How often delivery succeeds across a real building has not been measured.
Beyond Bluetooth range, delivery additionally depends on both phones having a working internet connection at the time, on the guardian's phone being permitted by its operating system to keep watching in the background, and on our relay being reachable. Mobile operating systems may stop background activity at any time, and we cannot prevent it. An alert that cannot be delivered within twenty minutes is discarded rather than delivered late, because a warning that arrives hours after the event invites action that is wrong.
Alerts sent beyond Bluetooth range are deliberately fewer. The detector must be substantially more confident before an alert is forwarded than before it is shown on a phone in the building. This is a measured trade, described in clause 1: it reduces false alarms to roughly one per school per year at the cost of never forwarding about two in five of the detections that would raise a local alert. Silence is not evidence that nothing happened. This app must not be relied on as a means of knowing your child is safe.
None of the above is under our control, and we do not warrant delivery.
What we are responsible for
We are responsible for our own deliberate wrongdoing, and for anything the law where you live does not permit us to exclude — which, in many places, includes death or personal injury caused by negligence.
Nothing in these terms takes away a right you have under consumer law.
What we are not responsible for
Subject to clause 10, we are not responsible for harm arising from the app failing to detect something, warning you about something that did not happen, warning you late, or failing to reach you at all. Nor are we responsible for decisions taken in response to an alert.
Do not drive to the school
If you are a guardian and you receive an alert, do not drive to the school. You will arrive into a situation you cannot assess, in a vehicle that may obstruct the people who can help. Call, and wait to be told what is happening.
Governing law
These terms are governed by the laws of the Republic of the Philippines. This does not remove any protection you have under the consumer law of the country you live in.
Contact
Llyr Labs — contact@shotdetect.com. See also our Privacy Policy.