How a modern VMS scales from one site to city wide camera estates: independent services, one camera pull for every viewer, smart retention and standby servers.
Many video management systems were designed when 100 cameras counted as a large site. Today, smart city programmes span tens of thousands of cameras, and global enterprises run cameras across hundreds of locations. Estates of that size need a platform designed for growth from day one, not one stretched beyond its original design.
Visylix is built as a set of independent services, covering streaming, AI, the interface and data, packaged as a Docker image. The heavy parts can be moved onto their own hardware as demand grows, so you add capacity exactly where it is needed instead of oversizing everything for peak load.
Visylix connects to cameras over RTSP, RTMP, SRT and ONVIF through its own native streaming engine, built for large camera counts on standard servers.
Each camera is pulled once. The same video then feeds every live viewer over WebRTC or HLS and the recorder at the same time. A hundred people watching one camera never means a hundred connections to that camera, which keeps cameras and networks calm as audiences grow.
AI and video conversion are the most demanding work in any surveillance system. Visylix avoids unnecessary conversion by passing H.264 and H.265 straight through wherever the viewer can play them.
AI runs on GPUs where they are available and on standard processors where they are not. How many cameras a server can analyse depends on which of the 22 AI analytics you run and how often, so Visylix is sized to your actual estate. Under heavy load, the AI engine protects itself and recovers on its own, so live video and recording carry on even if analytics are under strain.
At city scale, retention quickly becomes the largest cost. Visylix gives you control at every level. Recent footage lives on fast local storage, and older footage can be archived to your own S3 compatible storage.
Continuous, scheduled and event recording can be mixed per camera, retention is set per camera or group, and locked evidence is protected from deletion. Recording can switch to full quality automatically when AI detects an event, or when an operator asks. Locked evidence is sealed with SHA-256 and exports as a self contained player that anyone can check offline.
Surveillance that matters cannot simply stop. Visylix runs as a primary and standby pair, with operator controlled cutover in under 60 seconds, and backups you schedule on your own infrastructure. The operator always knows when a cutover has happened and why.
Every part is built to fail gracefully. If AI processing is interrupted, cameras keep recording and live view keeps running. An in product device health console shows the state of every camera and server, so problems are caught before they become outages.
Around this core sits the full Visylix control room: video walls up to 6×6, floor plan and geographic maps, an alarm and incident console, Sudarshan automation and the Radha AI copilot, all running on your own hardware.
Visylix is built from independent services for streaming, AI and data, so each can grow on its own hardware. Each camera is pulled once and relayed to every viewer and the recorder, which keeps cameras and networks efficient as estates and audiences grow. Capacity is sized to your cameras, resolutions and analytics.
It depends on which of the 22 analytics you run, how many frames per second each analyses and the GPU itself. Visylix sizes this with you against your real camera list, and AI can run on standard processors where a GPU is not available.
Recent footage sits on fast local storage, and older footage can be archived to your own S3 compatible storage. Retention is set per camera or group, recording can switch to full quality when AI detects an event, and locked evidence is protected from deletion.
Visylix runs as a primary and standby pair, with operator controlled cutover in under 60 seconds and backups scheduled on your own infrastructure. Components fail gracefully: if AI processing stops, cameras keep recording and live view keeps running.