And from the twenty six other sensors sitting in the same device. Vision, location, motion, sound, pressure, light, radios and tags, all on one edge node, all in one event stream, all yours to build on.
Every phone is already a sensor package with a GPU attached. Read all of it from one place, in one timeline, with one set of rules on top.
Wide, ultra wide and tele. Objects, counts, states, conditions and positions.
Distance to what you are looking at, and volume in the frame.
Position, speed, heading, altitude, geofence entry and dwell.
Motion, vibration, shock and impact, harsh braking, tilt.
Rotation rate and stable orientation in three axes.
Compass heading. Which way the asset is actually facing.
Pressure and altitude change, down to a floor or a lift.
Sound level and acoustic signature for alarms, leaks and machines.
Lux at the point of work. Shift changes, enclosures opening.
Something close to the device, and whether it is still mounted.
Carried or fixed, walking a round or sitting on a machine.
Device and battery temperature, reported continuously.
A trusted timestamp on every reading, aligned to your shifts.
Light the subject on demand so the camera reads in a dark bay.
Alert the operator at the point of work, right where it happens.
A live operator interface on the same device doing the sensing.
Access point identity and signal strength as an indoor position hint.
Carrier, generation, signal and throughput, logged with every event.
Nearby beacons and tags, so a reading knows which asset it belongs to.
Tap an asset or a badge to bind a reading to an identity on the spot.
Centimetre range to a tag on the devices that carry it.
Temperature, humidity and vibration tags paired straight to the device.
Engine hours, fuel, fault codes and odometer alongside the camera.
Scales, meters, scanners and controllers over USB on the go.
Clip on a thermal module and run heat detection next to vision.
Pair a survey grade receiver when you want tighter positioning.
Combine any inputs you like and Lightapp writes them into a single record, timestamped and ready for your systems.
The impact, the g-force, the place it happened and the clip, arriving together as one record.
What was seen, which asset it was, where the device stood and when, signed by the device itself.
Vibration and sound signature tracked every second, so a change shows up the moment it starts.
Inference starts when the device enters the yard and rests when it leaves, which stretches the battery across a full shift.
Which trailer, bin or machine it is, what condition it is in and how warm it is running.
Every device charged, connected, mounted and awake, confirmed before the shift starts.
A fleet of sensors, a server and a dashboard is what the telematics products in your industry are made of. You have all three here, on hardware you can buy anywhere on the planet.
Location, driving behaviour, arrivals, dwell and the footage that explains the moment.
Running, idle, changeover or down, with counts and cycle times on any equipment.
Zone awareness, PPE at the point of entry, and a timestamped record that stands up to an audit.
Check every unit and keep the frame that justified the call, with the light on when you need it.
What is on site, where it sits, how long it has been there and what condition it arrived in.
Every event in one store, so throughput, downtime and utilisation are queryable on demand.
Each of these was built by somebody who had the problem and wanted the number.
If a sensor can pick it up, you can turn it into a number, a threshold and an action.
Everything runs from the browser, on a phone you already own, starting the moment you open it.
One tap in the browser. Processing begins on the device immediately.
A general detector covers 80 common classes so the loop is live from the start.
Draw one box around your part. Examples are gathered as you move around it.
Location, motion, sound and radios join in, so every event carries its context.
Set the rule, point it at your webhook, and your own systems take it from there.
The infrastructure layer is finished and running in production. You start at the part that makes it your product.
Both are yours to drive. Pick per project, or start with one and promote it.
MobileNet embeddings with a kNN head, trained from a box you draw plus background crops gathered automatically. Enough to prove the idea in one sitting, running entirely on the device.
Record a clip, annotate it, send the dataset to train. A proper detector comes back exported and hosted, and promotes itself in place while the camera keeps running.
Here is exactly what runs on your device and what comes back out of it.
// one event, every stream that was live at the time { "app": "dock_throughput", "rule": "count_change", "class": "pallet", "n": 37, "confidence": 0.94, "device": "dock-04", "at": "2026-08-16T09:04:11Z", "location": { "lat": 41.9042, "lon": -85.6394, "speed": 0.0, "geofence": "yard-3" }, "motion": { "peak_g": 0.04, "heading": 184, "mounted": true }, "env": { "audio_db": 62, "lux": 340, "temp_c": 31.4 }, "tags": [{ "ble": "trailer-88", "rssi": -62 }], "snapshot": "https://…/snapshots/…jpg" }
Everything you make here is exportable from day one. Take the annotated dataset as a YOLO zip and train it anywhere. Take the weights and serve them yourself. Point the device at your own model. Pull the full event history over the API whenever you want it.
See the formats and the full stackimages/ labels/ data.yaml
model.onnx
REST, full history
register in models.js
Sensing and inference run on hardware you already own. Plans cover the assistant, the training GPU and how many devices you pair.
Cancel any time. Everything you make stays yours. Questions: info@lightapp.com
Pick up a phone and find the number. Four minutes from here to a live reading.