Connect
Point it at an ONVIF or RTSP stream from a camera you already own, or drop in a clip. Nothing is installed on the camera and nothing about its wiring changes.
The demo on this page runs the same loop on a clip, in your browser.
Your cameras alert all night about rain, headlights and branches. This picks out the ones that were a person, a vehicle, an animal or a package.
Interior, after hours
Fixed camera, low light, two subjects
Analysing the clip
Clip
Decoded in your browser. Your video is never uploaded.
Sets the luma delta threshold and the smallest region that counts.
Kills the one frame flicker that no operator wants paged about.
Exclude zone
Put it over the road, the tree, the neighbour's window. Events inside it stop alerting.
The frame an event was found on shows up here.
Hover a row in the event table to preview its frame, click it to hold the frame and jump the clip there.
0
Motion events found
0
Dropped by zone
0
Dropped, too short
0
Alerts you would get
| # | Start | Length | Class | Conf. | Area |
|---|
Reading the clip. Events appear here as they are found.
Get these alerts from your own cameras.
Rain, headlights, branches, shadows and a moth on the lens stop firing. What is left is the handful of events you would have wanted someone to look at.
Because it classifies the object, not the pixel change.
You get a list of events with timestamps, classes and durations, so you jump straight to 02:14 instead of scrubbing six hours of an empty yard.
Because every event carries the moment it started and how long it ran.
It connects to the ONVIF and RTSP cameras already on the wall. Hikvision, Dahua, Reolink, Axis, Amcrest, UniFi, Lorex. No new hardware, no installer, no rip and replace.
Because the thing that was broken was the software, not the camera.
The track runs sideways: swipe it, scroll it, or use the arrow keys once it has focus.
Point it at an ONVIF or RTSP stream from a camera you already own, or drop in a clip. Nothing is installed on the camera and nothing about its wiring changes.
The demo on this page runs the same loop on a clip, in your browser.
A background model of the empty scene is built from the stream, then every frame is compared against it. Sensor noise is shrunk away by an erosion pass and what survives is grouped into regions that moved together.
This is the stage your NVR already does, and it is the stage that fires at rain.
Each detected region is cropped at its strongest frame and classified: person, vehicle, animal, package, or other motion. The class, a one line description and a confidence value are attached to the event.
Categories only. No face templates, no gait, no re-identification, by design.
Events that survive your zones, your minimum duration and your class filter become the alert. Email on every plan, webhook and Slack from Operate up, schedules and escalation on Multi-Site.
Everything else stays in the timeline, searchable, without paging anyone.
This is the interior clip playing in the demo above, at sensitivity 6, with the minimum duration set to 1.5 seconds and one exclude zone drawn over the lit doorway on the left. Nothing here is illustrative: it is the output of the same code your browser just ran, captured and written down.
Run it yourself, or upload a clip of your own site: open the demo.
Captured 6 September 2026 on the bundled interior clip: 18 seconds of footage, 94 frames sampled, in a desktop browser. The classes came back from the model as person, on all eight, between 80% and 98% confidence. Your own run will not match this one to the event, because your machine samples the clip at its own rate. That is what it means for the numbers to be real.
Swipe the table sideways to see all four.
| Motiondetection | Enterprise camera platform | Your NVR's smart detection | Open source, self hosted | |
|---|---|---|---|---|
| What you actually buy | Software for the cameras on your wall | A whole camera system, cameras included | Included with the recorder | A licence of zero and a project |
| New hardware needed | None | Every camera replaced | None | A GPU box or an accelerator |
| Published price | Yes, on this page | Quote after a sales call | Included | Free, plus the hardware and your weekends |
| Per camera, per month | $8.32 to $12.25 | Roughly $100 and up | Nothing extra | Hardware and maintenance instead |
| Buy it without talking to anyone | Yes | No | Already bought | Yes |
| What it detects | Person, vehicle, animal, package, other | Broad object and behaviour analytics | Pixel change, with a person filter bolted on | Person, vehicle, animal, depends on the model |
| Face recognition | Never, by design | Usually a core feature | Sometimes, on newer models | Available if you add it |
| SSO, roles, audit log | From Multi-Site up | Yes | Local users on a box | You would build it |
| Where it beats us today | Live camera health and PTZ control are not part of it | Live camera health, PTZ control, a hardware warranty | Nothing to install, and it is already paid for | Full control, no subscription, no vendor |
Prices for the other three are their public positioning, not a quote we have seen. The last row is there because conceding the easy cases is the only way the hard one stays believable.
At the enterprise anchor of about $100 per camera, fifteen cameras is roughly $1,500 a month, most of it paying again for hardware you already own. Annual billing takes another 20 percent off.
One site, a handful of cameras
$49 per month
$39 a month billed annually
4 cameras
$12.25 per camera
The operations manager
$149 per month
$119 a month billed annually
15 cameras
$9.93 per camera
Multi-site facilities
$499 per month
$399 a month billed annually
60 cameras
$8.32 per camera
Unlimited cameras and sites, custom retention, SIEM and syslog delivery, an on-premise connector, a custom MSA, purchase order and invoice billing, and a named contact with an SLA.
Every plan is paid. The demo above is the free part, and it is the whole detection loop, ungated.
Motiondetection classifies object categories: person, vehicle, animal, package. It does not build face templates, it does not do face recognition, and it does not do gait or re-identification. That is a product decision, not a setting, so it cannot be switched on by accident in Illinois, Texas or Washington.
The person watching wants fewer alerts. You are buying the case where nobody was watching because the channel had been muted for a week. Those are the same purchase.
SAML and OIDC single sign on, roles for Viewer, Operator and Admin, and per site scoping so a contractor sees one yard.
Who viewed which clip, who changed which zone, who exported what, with a retention window you choose.
Encrypted in transit and at rest, a stated retention policy with configurable deletion, a DPA, and a published sub-processor list.
Annual invoicing, purchase orders, a W-9, and a security questionnaire answered rather than deflected.
At 15 cameras, an enterprise platform at about $100 per camera is roughly $18,000 a year, most of it paying for hardware you already own. Operate is $1,428 a year. Check it with your own camera count before you believe it.
Any camera that speaks ONVIF or RTSP, which is effectively every business IP camera sold in the last decade: Hikvision, Dahua, Reolink, Axis, Amcrest, UniFi, Lorex and the rest. If you can pull a stream URL out of it, it works.
No. That is the entire point of the product. The cameras stay, the recorder stays, the cabling stays. What changes is what reads the video.
No, and it never will. It classifies categories: person, vehicle, animal, package, other motion. There are no face templates, no gait analysis and no re-identification across cameras. This is what makes it defensible under the Illinois biometric law and its equivalents.
In the demo on this page, nowhere. Your clip is decoded by your own browser and never uploaded. Only small still crops around each detected event are sent to be classified, and only for the events the demo found. On a live deployment you choose the retention window and can delete on a schedule.
Your NVR mostly does pixel-difference detection with a person filter bolted on top, which is why rain and headlights set it off. Motiondetection classifies what the moving object is before anything alerts. To be fair to your recorder: for one camera pointed at a door in good light, the built-in detection is genuinely fine. It falls apart at four cameras outdoors at night, which is where this starts paying for itself.
Operate, $149 a month, covers 15 cameras, so twelve fits with room to grow. That is $9.93 per camera per month, or $119 a month billed annually.
Accurate enough that the alert channel becomes readable, and not accurate enough to be left alone with a fence line. It struggles with heavy rain directly on the dome, very low light with no working IR, and a subject that is mostly out of frame. We would rather name those than quote a percentage we cannot support. Every event carries a confidence value so you can see when the model was unsure.
You confirm your email with a 6 digit code and sign in with that address. Pick the tier that matches your camera count, pay by card in Stripe's own checkout, and the console opens: cameras, exclude zones, the classified event timeline and alerts. Nothing is charged until you pick a tier, and the demo on this page stays open and stays free.
One field, one 6 digit code, and your console opens on that address. Pick the tier that matches your camera count and pay by card, and the demo above stays open either way.
One field. We will send a 6 digit code to confirm it is you.
No card needed to sign up. Confirm the address, sign in to your console, and pick the tier that matches your camera count.