1. Home
  2. Products
  3. NSure AI-in-the-Box
NSure AI-in-the-Box

Video analytics AI
that finishes on site, no server

NSure AI-in-the-Box is a low-power Edge Video AI solution that analyzes CCTV video in real time inside a single on-site device, with no monitoring server or GPU server. Only Korean-language situation descriptions and event text are sent to the control center.

Input
4 CCTV channels (1080p 30fps, RTSP)
AI compute
NPU 48 TOPS (Qualcomm QCS8550)
Power
12V, 36W low power
Connectivity
Built-in 5G and e-Um 5G (n79) module
Overview

Video stays on site, only text goes to the control center

A lightweight AI keeps watch as the first stage, and when something suspicious is detected, an on-device VLM makes a second-stage judgment on context to filter out false alarms.

  • 01

    No monitoring server, no GPU server

    A single box handles multi-channel streaming analysis end to end.

  • 02

    Continuous first-stage detection, VLM second-stage judgment on events

    A second look at context sharply reduces false-alarm notifications.

  • 03

    Raw video never leaves the device

    Video is stored on the device; only situation descriptions and metadata are transmitted. Privacy and air-gapped network requirements are met by design.

  • 04

    12V, 36W low power, installs without new wiring

    Deploy anywhere: drones, vehicles, rail cars, tunnels, substations.

  • 05

    Built-in 5G and e-Um 5G (n79) module

    Stays connected to the control center on the move and inside private 5G networks.

Processing flow

Six steps from capture to control center notification

  1. Video capture

    Multi-channel CCTV RTSP

  2. Continuous detection

    Lightweight AI, first stage

  3. Event trigger

    Suspicious frame selection

  4. VLM context judgment

    Second‑stage false‑alarm check

  5. Situation description

    Korean-language summary

  6. Control center notification

    Text only

Platform

A processing pipeline that completes inside one device

Everything from 4-channel CCTV input to control center notification happens inside the device.

Input

4 CCTV cameras

  • 1080p 30fps, RTSP streams
Software

NSure Edge-AI Platform

  • Continuous detection engine (stage 1)8 lightweight detection models, 4 channels in real time on the NPU
  • Event triggerFrame selection and queuing
  • On-device VLM (stage 2)3B-class quantized (W8A8) model for context judgment and false-alarm verification
  • Event pipelineGenerates Korean-language situation descriptions and metadata; raw video stays on the device
Hardware

Pegatron Grizzly

  • Qualcomm QCS85504K decoding and ISP, Kryo 8-core CPU, NPU 48 TOPS (INT8), Adreno 740 GPU
Output

Control center dashboard

  • Receives events and situation descriptions, false-alarm review and statistics, LLM Assist PoC
Deployment sites

Stronger where the site is on the move

  • Construction sites

    Linked to drones and mobile CCTV

    Monitors worker safety and site access where only temporary power is available.

  • Roads · Tunnels

    Distributed sites with no server

    Detects stopped vehicles, wrong‑way driving and accidents on the spot.

  • Rail cars · Buses

    5G connectivity on the move

    Detects falls, violence and fire inside the cabin, even while in motion.

  • Manufacturing

    Linked to patrol robots and equipment CCTV

    Continuously monitors equipment anomalies and worker safety.

Specifications

Hardware specifications

A single Qualcomm QCS8550-based device analyzes 4 CCTV channels in real time. With 12V, 36W low power and a built-in 5G module, it installs anywhere without new wiring.

NSure AI-in-the-Box hardware specifications
Processor (SoC) Qualcomm QCS8550
AI compute (NPU) 48 TOPS
Memory 16GB LPDDR5x
Video processing 4K decoding
Network 5G, 4G, Wi-Fi 7
Interfaces RS-232 / RS-485
Power 12V / 3A

Want to see NSure AI-in-the-Box
running on site?

A free IP visibility assessment is not a big project. We agree the scope and timeframe, and you get numbers you can report to management.
Request a consultation and we will get back to you within 24 hours.