Ten modules, one platform, one live wall — each running on the feeds you already have, each drawing as an overlay on live and recorded video.

Detects and reads number plates across multiple regional formats, corrects misreads and validates against real state codes, then enriches each read with vehicle type + dominant colour and — optionally — owner details via an RC/VAHAN lookup. Entry and exit reads pair into journeys.
Hard-hat / helmet detection with a bare-headed-worker check, gated to the zones where PPE is actually required so you get signal, not noise.
Slot-occupancy detection fused with ANPR into a single parking session: which vehicle sits in which slot, for how long, with overstay alerts.
Triple-riding, no-parking-zone entry, wrong-way and over-speed detection — each violation named to the offending plate and owner.

Enrol employees from reference photos, run watchlists, and blur faces for privacy — architected to stay on-prem and licence-clean.
Firearm and edged-weapon detection with a two-step confirm before a critical alert fires — tuned to minimise false alarms.

Detect and count people and objects, estimate demographics, draw intrusion and counting zones, and build occupancy heatmaps.

24/7 continuous recording with motion-aware storage tiering, plus fast indexed forensic search across the whole archive.
On-device object & gender detection that runs entirely on the phone, plus "Go Live" — turn any phone into a camera that streams straight to the wall over a secure relay.
Every event fans out to webhooks, email and gate/barrier controllers in real time.
The number-plate recogniser runs on-prem across regional formats, corrects misreads and validates real state codes — then pairs each read with owner and vehicle details.

Plate text + region above are live output from the on-prem recogniser. Owner & vehicle fields are shown illustratively — a vehicle-facing gate camera supplies them on site.
A media gateway pulls one warm stream per camera from your NVRs — shared by the live wall, recorder and every AI module, so the NVR only ever serves one session.
Licence-clean AI models run in real time on-prem — CPU-only where needed — writing detections that overlay live and recorded video alike.
Named notifications, gate triggers and webhooks fire instantly; continuous recording + forensic search keep the evidence.
No. VisionVault is self-hosted — every frame and every model runs on hardware you own. There are zero cloud dependencies, which is exactly why privacy-sensitive sites choose it.
No. It works with your existing IP cameras and NVRs over standard IP. No rip-and-replace.
Every model is trained on commercially-usable data with verifiable provenance — no non-commercial, GPL or unverifiable weights. That means you can deploy it commercially without licensing risk.
Yes. Core detection, ANPR and PPE run CPU-only; heavier workloads scale with a GPU when you have one.
Every event is published as a notification that auto-fans to webhooks, email and gate/barrier controllers — so you can wire VisionVault into what you already run.
We'll set up a live demo on cameras like yours and walk you through the modules that matter to your site.
Request a demo →