BongoSense — Intelligent Retail Security

See It In Action

Eyes on every aisle. Brains in every camera.

Same CCTV you already have, with a layer of AI that explains what it sees.

app.bongosense.com

BongoSense dashboard — operations overview

Live ops dashboard

Every event, alert, parking and occupancy figure for all your sites in one feed. Tablet-friendly for guard stations.

edge · live frame

Edge camera view — pose tracking with per-person IDs

Pose tracking + per-person IDs

The edge tags every person on screen with a tracking ID and skeleton. The vision-language model uses that context to give each event a clear YES/NO verdict.

What We Offer

One platform. Eight pairs of eyes.

Pick what each camera watches for — theft, falls, queues, parking, occupancy or tampering. Same hardware, same dashboard.

🛡️

VLM Theft Detection

A vision-language model watches every shopper's behaviour and gives a clear YES / NO verdict with a one-line reason in plain English — "Subject is concealing a bottle inside their jacket." No more cryptic anomaly scores to argue about.

⛑️

Slip & Fall Detection

Pose tracking spots when a customer goes down — a critical signal for liability and rapid response. Surfaces in the live alerts feed within seconds, with the camera, timestamp and track ID.

🚗

Vehicle Counting & Parking

Know exactly how busy your lot is at any moment. Live "currently in lot" count, today's entries and exits, and an hour-by-hour bar chart showing your peak inflow.

👥

People Counting & Occupancy

Tripwire-based entry / exit counters per door, rolled up to a single live store occupancy figure. Daily auto-reset in your local timezone — or zero it on demand.

🧾

Queue & Wait Monitoring

Draw a zone over each till, set a max length and max wait time, and get an alert the moment the queue exceeds it. Per-lane tracking — staff knows exactly which till to open next.

⚠️

Tamper Detection

If a camera is covered, knocked or repointed, you'll know before the next shift. One-click baseline reset from the dashboard once it's back where it belongs.

📱

Instant Telegram Alerts

The moment something fires, your team gets a Telegram message with the camera, timestamp, the verdict reason, and a thumbnail. Pause and resume per-recipient — no need to silence the whole site.

🎬

Evidence Clips & Review

Every theft alert captures a clip from your main stream. Tinder-style review queue lets your team mark events true or false positive in seconds — and that feedback goes straight back into model training.

📊

Live Dashboard

All your sites, cameras and events in one place. Filter by date, camera or outcome. Tablet-optimised live alerts feed for guard stations.

🏪

Multi-Site Ready

Manage multiple store locations from a single login. Each site has its own alert contacts, quiet hours, operating hours, and camera configuration.

🔒

Secure by Design

Your footage stays yours. Detection runs on a local edge server — no live feed leaves your premises. Evidence clips are stored encrypted with time-limited access links, isolated per client.

📜

DPA-Compliant

Built to meet Botswana's Data Protection Act, 2018. Auto-purge of unreviewed events after 3 days, explicit DPA notice on clip downloads, and audit trails on every action.

The Process

Simple to set up. Powerful in action.

We handle the technical side. You get the results.

We install a local server at your site

A small, quiet device connects to your existing CCTV cameras. It runs the AI detection model locally — no internet required for detection, no lag, no feed leaving your premises.

You decide what each camera watches for

From the dashboard, toggle VLM theft detection, vehicle counting, people counting, queue zones, dead zones or fall detection — per camera. The same hardware does double duty: a till camera can monitor queue wait and theft at the same time.

The vision-language model gives clear verdicts

Instead of cryptic anomaly scores, every theft trigger gets a YES / NO verdict from a vision-language model along with a one-line reason — "Subject is concealing a bottle in their jacket while looking around suspiciously." Your team sees exactly why an event was flagged.

Your team gets a Telegram alert within seconds

Your designated contacts receive an alert with the camera, timestamp, the model's reason, and a thumbnail. Quiet hours and per-recipient pause built in — no waking the boss at 3am.

Review, label, and act from your dashboard

Log into app.bongosense.com, watch the clip, swipe true or false positive in the review queue. Every label improves the model — your data trains your detection. Live alerts feed, parking and occupancy dashboards, and a full VLM verdict log are one click away.

Stop Losses.See Everything.

Stop Losses.
See Everything.