AI Video Analytics Box - one edge box, 40+ algorithms

An AI video analytics box is an appliance that sits next to your network video recorder or switch, reads the video streams from your existing cameras and runs computer-vision algorithms on them - face recognition, people counting, intrusion, PPE checks, ANPR and so on. It is not a camera and it is not a cloud account: it is a small computer with an AI accelerator that adds intelligence to the CCTV system you already have, and it pushes the resulting alarms to your VMS or control room.

No. Any ONVIF or RTSP camera will work, at 2MP to 8MP and in H.264 or H.265. Analogue cameras can be brought in through your existing DVR or encoder. In a typical project we change nothing on the camera side - we add the box, re-point the streams in the console and start assigning algorithms.

From 4 to 64 channels on the same platform. As a rule of thumb, use one channel per camera stream and choose the model by camera count: XCORE-4 for up to 4 cameras, XCORE-8 for 8, XCORE-16 for 16, and so on up to XCORE-64. If you plan to run several algorithms on the same channel, or to record all channels at 4MP continuously, tell us and we will size the box with headroom.

Far less than a GPU server. The 16-channel edge box draws about 5 W at full load with a completely fanless design, and the larger 96-channel edge server stays under 100 W. That typically saves over 3,000 RMB in electricity per box per year compared with a GPU-based PC, and it means the box can live inside a cabinet in a dusty or remote site without fans to clog or filters to clean.

An AI NVR is a recorder with some detectors built in, so you are tied to that vendor's cameras, firmware and algorithm list. An edge AI box is an open analytics layer: it takes the cameras you already own, can run dozens of algorithms simultaneously, and pushes events to whatever VMS or platform you use. Many customers run the box alongside their existing NVR instead of replacing it.

No. All inference runs on the device and the alarms are generated locally, so the box keeps working with no internet connection at all. The face library, detection rules and recorded evidence are stored on the box. Cloud or central VMS is optional and is normally used only to manage many sites from one console.

Price depends on the channel model and the software bundle rather than on an annual licence - there is no per-camera subscription. Smaller models are the entry point for a single site; the higher channel models cost more but replace several smaller boxes. Send us your camera count and the algorithms you need and we will come back with a configuration and a quotation rather than a list price.

Yes, and we prefer it. Send a sample clip or camera stream plus the behaviour you want to detect, and we will run it through the box and return the annotated result. It is the fastest way to confirm accuracy before an order, and it is free.

Yes. We supply the appliance under your brand with your logo, packaging and boot screen, and we develop custom detection models - sample collection, training, validation and edge deployment - when your scenario is not covered by the existing library of 47 algorithms.