AI vision for casino tables
Owl AI watches every table, every hand, every second. It flags drowsy dealers, posture breaches and fraud as they happen. Run it locally on your own system, where nothing leaves your building, or let us run it in the cloud.
Local or Cloud. With Local, your data never leaves your building. →01 · Local or Cloud
Run Owl AI entirely on your own system, where nothing ever leaves your building, or let us run the AI in the cloud. The detectors, the Signals feed and the dashboard are the same either way.
Video and AI stay inside your networkVideo is sent to Owl AI Cloud for analysis
Your cameras, your PCs, your data. The AI runs on your own computers, and not a single frame of video ever leaves your building.
No AI hardware to buy or look after. Connect your cameras and Owl AI's cloud does the watching.
| Local On your system | Cloud Managed by Owl AI | |
|---|---|---|
| Where the AI runs | Your own PCs and servers | Owl AI's cloud servers |
| Video leaves your building | Never | Yes, sent for analysis |
| Footage & evidence stored | On your system only | Handled by Owl AI Cloud |
| Internet required | No (offline licence available) | Yes |
| On-site hardware | Windows PCs, NVIDIA GPU recommended | Cameras and an internet connection |
| Best for | Maximum privacy and control | Fast start, minimal hardware |
02 · The detectors
Each detector has a single job and does it in real time, from a live camera, a stream or an uploaded video. Scroll to watch each one work.
Real footage, real detections. Every landmark, eye ratio and timer is computed from licensed stock video by the detection engines, not animated. These dealers are working normally, so the drowsiness and posture feeds truthfully read all clear. The card feed then cuts to a clip where the dwell alarm genuinely fires.
Owl AI tracks the dealer's face and measures how open the eyes are, frame by frame. A blink is ignored. If the eyes stay shut for longer than the duration you set, a red signal fires, and a close-up of the face is pinned to the trigger.
Body and hand landmarks show what the dealer is doing: hands on the table, or a wrist raised to the ear. When an out-of-position posture lasts longer than the configured duration, Owl AI saves the exact frame and raises a signal.
Owl AI follows every card on the felt and every hand above it, through the round from deal to draw. A hand that lingers over a card zone, a face-down card where it shouldn't be, or a card out of sequence raises a red signal with the analysed frame attached.
03 · Signals
Every detector, table and worker PC reports into one feed. Only raised rules appear, never frame logs or pre-threshold blips. This is the real Signals console, running here with simulated alarms. Try the filters, open footage, close a signal.
Detection alarms, table pauses, and alerts for failed or stopped detectors.
Alarms are received even while you use another page.
| Received | Detector / table | Signal | Details | Actions |
|---|
Detection alarms and health incidents are saved. Health recovery closes its signal automatically.
04 · Getting started
Double-click to open the setup wizard. It checks the PC, installs what's missing and verifies your GPU.
Hub, Worker, or both on one machine. The choice is remembered for every later launch.
Connect each camera or stream and draw its detection area. Owl AI assigns tables to the worker PCs.
Signals arrive on the dashboard, the supervisor portal, Slack, Teams or email.
05 · Local architecture
This is how the Local option runs on your premises. The hub runs the dashboard, accounts, scheduling, health and notifications. Workers run detection and keep evidence local. The same package installs everywhere, so you can scale from one PC to a whole floor.
06 · Security & privacy
Permissions are checked by the hub on every request. Hiding a button is not the access control. Choose a role to see what it can open.
With Local deployment, video inference runs on your own PCs. Recordings and evidence remain on the worker that captured them, and nothing is uploaded for processing.
Live FPS, latency and heartbeats, with five states that are never blurred together.
A separate HTTPS listener serves only sign-in, assigned signals and permitted snapshots. It uses a private certificate authority that can't vouch for internet sites.
Choose which events go to which destination, filtered by table.
Ed25519-signed licenses, verified locally. Choose an online subscription or an offline license with no licensing server at all.
Account creation, permission changes and disabled accounts are logged with the actor and a timestamp. Passwords and tokens never are.
Fraud detection runs on NVIDIA CUDA. Setup checks your driver, VRAM and model inference before the first table goes live.
Snapshots are taken at the trigger, not around the clock. Each alarm keeps its exact frame, rule and timestamp.
07 · Pricing
Your license sets a table allowance, enforced across all your worker PCs. Choose how you'd like to pay.
10 tables · billed monthly
Price on request
10 tables · fixed term
Price on request
08 · Questions
Not with Local. Detection runs on your own PCs, recordings and evidence stay on your system, and only the alert messages you configure go out to Slack, Teams or email. With Cloud, video is sent to Owl AI Cloud for analysis, so you need no AI hardware on site. You choose which model fits your policies.
Choose Local if privacy, compliance or data residency come first: everything is processed and stored on your own system, and it can run without internet. Choose Cloud if you want to start quickly without buying or maintaining AI hardware. The detectors, Signals feed and dashboard are the same in both.
With Cloud, just your cameras and an internet connection. With Local: Windows PCs. The installer sets up its own Python runtime and checks every requirement. Fraud detection uses an NVIDIA GPU via CUDA. Drowsiness and posture run on CPU or on the browser GPU. One strong PC can watch several tables, or you can give each table its own PC.
Live cameras, video streams and uploaded recordings. Uploaded videos play back at normal speed, and each alarm links to its exact timestamp.
Not by default. Supervisors see signals for their assigned tables only, plus alarm snapshots if you allow it. Live streams and full recordings stay with Management and Administrators.
Cloud needs a connection. Local with an offline licence doesn't: it is verified on your system and never phones home. Local subscriptions check payment status periodically and keep working for up to 24 hours without a connection.
Book a walkthrough on your own footage. We'll show you what Owl AI would have caught.