Client engagement · Live demo

Watching 100 feeds is impossible.

What if you could just describe what you're looking for, in English?

OmniSight is an AI-assisted surveillance system: write a rule in plain English, and it watches every feed for it.

Why this is hard

More cameras, more guards, more motion alerts — none of it scales.

Attention doesn't scale

Human focus breaks down past ~3 feeds. A wall of monitors becomes background noise within an hour.

Motion alerts cry wolf

Standard motion sensors fire on every shadow, leaf, and headlight. Real incidents get buried under false alarms.

The approach

Describe → Detect → Alert.

Instead of training a model for every scenario, OmniSight lets you describe what you're looking for in plain English. AI vision checks each frame for it.

01

Describe

Write the rule in plain English — like "Person in restricted area after 6pm" or "Vehicle blocking the emergency exit". No special syntax. No new model to train.

02

Detect

AI vision checks each frame from your camera against the rule you wrote. When it finds a match, faces in the frame are saved for the review gallery.

03

Alert

When the AI confirms a match, you get an alert with the frame attached — in seconds, not minutes.

Built with

The interesting part was figuring out how much could run in the browser before needing a server.

Next.js 16React 19face-api.jsMediaPipeONNX RuntimeGPT-4o visionTailwind 4

Run the demo on your own webcam

Two-minute live session. Type a rule, the system watches the feed for it, you get the alerts back at the end.

Start the demo
Honest scope

This live demo runs on a single webcam to show the core idea end-to-end. A full production rollout would add many cameras working together, a database, user permissions, data-retention rules, and a faster server-side AI step.