Sauron
Sauron turns a camera estate into security intelligence your team can review and act on. It detects and correlates events such as loitering, tailgating, perimeter breaches and camera tampering, then escalates them to an operator. Processing happens on your own premises, so raw footage and biometric data stay inside the building.
Person detection and dwell-time tracking across camera views. Face matching against an enrolled gallery of staff and authorised visitors. Optical fault detection identifies tampering, glare and obstruction, and separates gradual drift from sudden change.
An incident engine groups related events instead of presenting a flat alert feed. An investigation workspace follows one question across multiple events and sites. Structured search covers events, incidents, investigations and device faults.
Consequential actions wait in an approval queue until a person releases them. Seven roles run from owner down to read-only viewer. Every event carries a review workflow, and access to evidence is itself written to the audit log.
Edge gateways buffer detections locally through a network outage and deliver them in order once the link returns. A detection is written to local disk before any network attempt. Ingest is idempotent, so a retry after a lost response is a replay rather than a duplicate.
Event and incident volume, response times, device availability and the most common event types. Reporting also ranks which sites generate the most recurring problems, worst first, so attention goes where it is needed.
No model in the cloud
The central instance refuses raw camera intake and never loads a recognition model.
AES-256-GCM
Biometric data and evidence are encrypted at rest. TLS protects data in transit.
Key rotation
Every encrypted item is key-tagged, so keys retire and data re-encrypts without downtime.
Tenant isolation
Enforced structurally rather than by a query parameter, with abuse protection on login and signup.
The same software runs in two places. At your site it does the actual work: person detection, dwell-time tracking, face matching against a locally enrolled gallery, and optical tamper and glare detection. In a central instance it receives only finished events, meaning what happened, where, when and with what confidence, and correlates those events into incidents across sites.
Raw camera frames, face embeddings, movement traces and evidence files are structurally incapable of crossing that boundary. This is not a policy setting or a configuration option. The central instance refuses raw camera intake outright and never loads a recognition model at all. A misconfigured device, a copied credential or a well-meaning engineer cannot open that door, because the door does not exist on that half of the system. You can adopt AI security analytics without your footage becoming someone else’s asset.
There are two deployment models. Local plus central is the recommended one: cameras feed a Sauron server on your premises, on-site staff work from a local dashboard, and a one-way uplink carries derived events to a central instance for oversight across sites. Fully local is the same deployment with the uplink switched off, so nothing leaves the building at all. That suits sites which cannot be networked outward. The software is identical in both cases. There is no cut-down edition and no feature held back for a higher tier.
Sauron suits any organisation accountable for what its cameras saw. That includes industrial plants and utilities, commercial property and retail estates, logistics depots and transport hubs, healthcare and education campuses, and government facilities. It is a particularly close fit wherever a regulatory or contractual obligation prevents footage being handed to a third party’s cloud.
Tell us about your estate, the systems already in place, and what you need to see. Leave a contact address and we will come back to you.