Where Inference Runs Is a Deployment Choice.
Inference runs on GPU hardware — either on-premise at each location, on GPU-equipped edge devices, for the lowest possible latency; or centrally in the cloud, fed by an on-site VMS (Video Management System) that streams camera feeds up for customers who would rather not invest in edge GPU hardware at every location. Which model fits is a deployment decision made per customer, not a fixed architecture.
“Where inference runs is a deployment decision, not a fixed architecture — latency-critical actions favor the edge, lower upfront hardware investment favors the cloud.”
Two Deployment Models
Not Every Site Gets the Same Architecture
Some customers want GPU hardware at every location. Others would rather not — and stream video to the cloud instead. Proglint supports both, chosen per customer.
GPU-equipped hardware installed at the location itself runs inference directly against the live camera stream — no round trip to a distant server for a real-time decision.
- Detection and the resulting decision happen with minimal delay — no waiting on a network round trip.
- Real-time actions (an instant alert, a PLC reject signal) depend on local inference; a network hiccup should not stall a safety or quality decision.
- A good fit when the customer wants to invest in GPU hardware at each location for the lowest possible latency.
A lightweight VMS (Video Management System) is deployed on-site instead of GPU hardware. It streams the camera feeds to a cloud-hosted GPU inference engine, which runs the same detection pipeline centrally.
- No GPU hardware to provision, rack, or maintain at each individual location — only a lightweight on-site VMS streaming the feed.
- Faster rollout across many locations at once, since there is no per-site hardware procurement or install to coordinate.
- A good fit when the customer would rather not invest in edge GPU hardware everywhere, and the use case can tolerate a small added latency.
The honest version: this is not Proglint choosing an architecture for the customer — it's the customer choosing how much GPU hardware investment they want at each site, and Proglint's platform supporting either choice on the same downstream pipeline.
The Edge Stack
What Runs on an Edge Device, When That's the Chosen Model
Three components, working together on-site, before anything reaches a wider network.
NVIDIA DeepStream SDK
Runs the GPU-accelerated inference pipeline — decoding, batching, and running detection models against the live stream, wherever the GPU lives.
MediaMTX (RTSP)
Handles RTSP stream ingestion from mixed camera vendors, normalizing feeds before they reach the inference pipeline.
Edge GPU Inference Servers
GPU-equipped edge devices deployed on-site — the physical hardware behind the on-premise deployment model.
Metadata-First Architecture
Only structured JSON metadata — bounding boxes, OCR text, classifications, timestamps — leaves the edge device by default. Raw video stays local unless specifically retrieved for an evidence clip, keeping bandwidth and data exposure minimal.
“Send the enterprise what it needs to act, not everything the camera recorded.”
Raw video footage. Whether inference is on an edge device or in the cloud, raw footage isn't the default payload leaving that system — it stays put unless specifically retrieved to attach an evidence clip to a flagged event.
Structured JSON — bounding boxes, OCR text, classifications, timestamps — sent via a Node.js REST API to the Enterprise Command Center, regardless of which deployment model produced it.
Constant Across Both Models
What the Cloud Always Handles
Whether a given site runs edge GPU or cloud GPU inference, these functions live in the cloud either way.
- Centralized management of the fleet — every location, camera, and device visible from one place.
- Multi-location visibility: aggregating events across every site into the Enterprise Command Center.
- Model management — deploying updated detection models across the network as they improve.
Edge and cloud GPU inference are both infrastructure underneath the Vision AI Engine’s detection models and the streams handled by Computer Vision Infrastructure. The metadata either produces surfaces in the Enterprise Command Center.
Ask About the Right Deployment Model for Your Sites
Schedule a customized demonstration with our enterprise AI engineering team. Discover how Proglint deploys across 2,500+ locations with zero camera teardowns.