Describe what to watch for. MATA does the watching.
MATA AI turns any camera into an AI that understands your rules — written in plain language — and alerts you the moment they're broken. No model training. No integration project.
Your cameras record everything
and notice nothing.
CCTV is passive — hours of footage nobody watches. You only look after something has already gone wrong.
Nobody is watching
Managers can't be everywhere. SOP slips, hygiene lapses and unmanned counters go unseen until a complaint or audit.
Legacy analytics is rigid
Traditional video analytics needs a per-rule ML model, a vendor integration project and weeks of tuning for every new thing.
Add a rule, call the vendor
Want to detect one more thing? You're back in a procurement cycle. The system can't adapt at the speed your operation changes.
From a camera feed to an alert
in four moves.
Connect a camera
Browser cam, RTSP, MJPEG, HTTP-snapshot, push feed — or upload recorded video.
Describe what matters
Write trigger conditions in plain language. Set severity and an alert policy. Live in seconds.
MATA watches every frame
A vision LLM evaluates each sampled frame against your rules, grounded in the auto-identified scene.
Alerts, evidence & reports
Real-time alerts with annotated snapshots, on-demand digests and structured offline reports — to Telegram if you like.
Type a rule. Watch MATA
compile it into vision.
You write one sentence. Behind the scenes, MATA auto-expands it (HyDE-style) into a precise visual checklist that grounds detection — you never write the technical part.
Everything you need to turn
cameras into awareness.
Alert policies that send signal,
not spam.
Per-trigger control over when a rule fires — so a busy kitchen doesn't become a wall of notifications.
Immediate
Fires on every detection, cooldown-gated so you're never flooded.
Repeated
N occurrences within M minutes roll up into one aggregated alert.
Persist
For ongoing states: fires on start, re-reports with elapsed duration, announces when it clears.
Continuous
Only fires once a condition holds for X seconds — tolerates a dropped frame.
One platform. Every
operation with a camera.
Restaurants are the flagship — but the same plain-language engine adapts to any environment.
From a camera that watches
to an operations brain that acts.
MATA isn't a camera gadget — it's the operations intelligence layer for any business with cameras. Here's where it's heading.
Now. Next. Later.
Everything is tagged so you always know what's shipping versus what's coming. Buyers trust honesty.
Operational awareness,
not covert surveillance.
Run MATA on your own machine or LAN. Point it at a local model so frames never leave the premises. This is accountability and business intelligence — not biometric ID, and never identity matching.
Login-protected
Form login with stateless HMAC-signed session cookies. Rotate the secret to log everyone out.
Local-model option
Point at a self-hosted vLLM backend — a zero-code switch — so frames stay on-site.
HTTPS + tunnel
TLS required; expose publicly via Cloudflare Tunnel behind login — no open ports.
Credentials never logged
Camera & stream URLs stored locally, masked in the UI, kept out of server logs.
The questions buyers ask first.
See MATA watch your
camera in minutes.
Describe a rule. Connect a feed. Watch it fire — with annotated evidence and a Telegram ping. That's the whole demo.