Every video analytics tool can tell you there's a person in the frame. None of them know whether that person was supposed to be there, whether the step they skipped mattered, or whether the inspection happened at all. Cyrus does — because it's built as a supervisor, not a detector.
Two kinds of tools are sold to sites today: video analytics, which detects objects and events in the picture, and data platforms, which chart telemetry on dashboards. Here's what each one can answer.
| Can it… | Video analytics Detection tools |
Data platforms Telemetry & dashboards |
Cyrus Operations Supervisor |
|---|---|---|---|
| Know how the work is supposed to happen | Detects objects & events | ~ Tracks numbers, not work | Procedures, schedules & rules, in your words |
| Notice a skipped step or a missed inspection | No | No | Flagged within seconds |
| Take on a new task | Months: collect data, train a model | ~ An integration project | Minutes, described in words |
| Write the report | Clips and counts | ~ Charts, not reports | Audit-ready; you check & save |
| Answer "what happened in bay 3 last Tuesday?" | Scrub the footage yourself | ~ If someone writes the query | Ask in plain language, get the moment |
| Remember your site | No retained context | ~ Data history, no judgement | Context, history & policy retained |
| Run on what you already have | ~ Often hardware-locked | Months of integration | Existing cameras & sensors, deploys in days |
| Get better with your feedback | Static after training | No feedback loop | Corrections compound, on your site |
Detection tools tell you what's in the picture. Cyrus tells you whether the work happened the way it should.
Detection tools were built to answer "what is in this frame?" Making them answer "did the work happen the way it should?" means rebuilding their core — slow, expensive and risky while Cyrus ships.
Cyrus was designed from the first line of code around knowing your procedures, rules and site — not bolted onto a pre-AI detection stack.
A new task is described, not trained. That's why a new procedure takes minutes, and why Cyrus rides every AI advance instead of ageing with one model.
Site memory and your team's corrections compound, so every deployment gets sharper — and the lead widens with each site you add.
For the technical reader. Frontier AI is powerful but expensive to run on live video; Cyrus's edge preprocessing and smart model routing make it viable in real time — measured against specialized small models and a direct frontier-model wrapper on real CCTV footage.
Internal benchmarks on real CCTV footage (altercations, firearm possession, industrial sites, conference venues), versus a commercial small VLM and a direct frontier-model wrapper. Results vary by deployment.
Each of these shifted recently — which is why nobody could build Cyrus before, and why the window is open now.
Today's models understand what's happening in a picture — not just which objects are in it — with no per-site training.
The cost of AI inference has fallen far enough that watching every camera and sensor, around the clock, is economically viable.
Most sites already run cameras and sensors — and won't pay for a rip-and-replace hardware refresh.
Memory, planning and tool use have matured enough to turn passive detection into a supervisor that alerts, records and answers.
The fastest way to understand the gap is to watch Cyrus supervise one procedure, side by side with what you have today.