How PARVEK fits around what you already have
Two patterns from live deployments. One spreads edge AI across dozens of small sites reporting into a command center the customer owns; the other adds a single appliance to a large site without touching its recorder.
Deployment patterns from live installations, with customer names, locations and commercial terms removed. Your site will differ — these show how PARVEK fits around infrastructure that already exists.
Multi-site retail chain
Edge AI at every site, with only summarized events reaching a command center inside the customer’s own HQ. No continuous video leaves any site.
Dozens of small sites, each with a handful of cameras and a domestic-grade internet link, plus an HQ team that needed comparable numbers across all of them. Backhauling video was never viable — the links could not carry it, and the customer would not accept continuous footage leaving each site.
- Footfall counting and dwell analytics
- Crowd and people monitoring
- Intruder alerts during designated hours
- Unauthorized access alerts
- Watchlist monitoring
- Early fire and smoke detection
- Camera tamper and camera health
- Staff attendance
- Alerts and events
- Alert snapshots and short clips
- Counts and analytics
- Device health
- Continuous video streams
- Full-time recordings
Design decisions
Edge AI per site, not centralized
Each site analyzes its own video locally, so detection keeps working when the link drops and no site depends on the others.
No continuous recording at the sites
Only short clips around alerts are retained. Live viewing stayed on the customer’s existing recorder app rather than being rebuilt.
Command center on the customer’s own server
The central store sits inside their HQ network, encrypted at rest, owned and controlled by them — not hosted by us.
Outbound-only connections
Sites and HQ open outbound tunnels; no inbound ports are exposed for remote users, and HQ accepts connections only from allow-listed site addresses.
Role-based access by site
HQ sees every site — alerts, clips, analytics and device health. Site staff see only their own site.
Process manufacturing plant
One on-premise appliance alongside the existing recorder, running three AI pipelines on cameras that were already installed. Recording was left exactly as it was.
A working plant with an established CCTV estate and a recorder the team relied on. Nothing about the existing setup could be disturbed, and the site needed gate vehicle logging, fire and smoke cover across critical zones, and people counting at the main gate.
- Vehicle entry and exit with plate recognition and counting
- Fire and smoke detection across critical zones
- People entry and exit counting at the main gate
- Camera health and tamper monitoring
- Optional known / unknown person identification at the gate
- Event logs
- Evidence clips
- Counts and reports
- Camera health
- Video to any cloud for analysis
- Data stored on the remote-access relay
Design decisions
No recorder replacement
The existing recorder continued handling video recording exactly as before. The appliance read camera streams directly or through it, whichever suited the network.
Existing cameras reused where they qualified
The existing estate carried fire, smoke and people counting. Only the plate-capture lane needed a camera specified for the job — fast shutter, correct angle, enough pixels across the plate.
AI local, no cloud dependency
All analytics ran on the on-premise appliance. Remote access went through an encrypted pass-through relay that stores no video or data.
No inbound ports
The appliance is never exposed to the internet. No port forwarding, no public IP; push alerts carry only the event summary.
Fire detection as a supplementary layer
Explicitly positioned alongside the statutory fire detection and suppression systems, never as a replacement for them.
Neither of these is your site
They are starting points. The assessment establishes which pattern fits, what your existing infrastructure can carry, and where something genuinely has to change.
Frequently Asked Questions
Do we have to replace our recorder?
No. In the single-site pattern above, the existing recorder kept handling video recording exactly as before, and the appliance read camera streams either directly or through it. Where a site wants one box for both, the appliance can also record — but that is a choice, not a requirement.
What actually leaves the site?
In both patterns, no continuous video. The multi-site deployment syncs only alerts, snapshots, short clips, counts and device health to the central server. The single-site deployment sends push alerts carrying just the event summary, with everything else staying on the appliance.
Do you need inbound firewall access?
No. Both patterns use outbound connections only — no port forwarding, no public IP, and the appliance is never exposed to the internet. Remote access runs through an encrypted pass-through relay that stores nothing.
Where does the central data live on a multi-site deployment?
On a server inside the customer’s own premises, encrypted at rest, owned and controlled by them. We do not host it.
Can a site keep working if its internet link drops?
Yes. Analytics run locally, so detection, local response and event logging continue. Data is buffered and syncs when the link returns. What stops is remote notification, which needs a network path by definition.
AI surveillance and analytics are technology-assisted and do not guarantee prevention or detection of every security incident or 100% accuracy.