Site Intelligence for Smarter, Safer Operations

Yes. Event detection, blurring, and encryption happen locally on edge devices, and raw CCTV footage does not stream off-site. Relevant clips are anonymised before they appear in cloud dashboards. Protex AI is designed to surface site patterns and process deviations, not to evaluate, score, or discipline individual workers.

Automation executes a defined task using fixed logic. Industrial AI identifies patterns, exceptions, and root causes across the wider site that a fixed process cannot see, such as recurring congestion or shift-level variation. The two are complementary: Industrial AI shows where and why performance changes, so teams know what to improve.

Site intelligence is the data layer that converts operational complexity into clear, immediate action. Protex Intelligence fuses every signal your site already generates, from cameras, systems of record, and equipment sensors, into one connected view, then finds the patterns and delivers leading indicators, root cause context, and recommended actions to every team.

Protex AI detects spills, walkway obstructions, clutter, misplaced materials, blocked emergency exits, and package jams on conveyor belts. When a hazard is detected, it automatically generates an event with a video clip. That event can't be marked as resolved until a corrective action has been documented, which closes the loop on remediation.

Detection to signal takes 1–2 seconds for real-time use cases. Because everything runs on the edge device rather than in the cloud, performance doesn't depend on network conditions. The system can also feed directly into PLCs and safety hardware to trigger physical interventions like machine stops or automated warnings.

A few ways. Real-time alerting rules require 99% precision before they go live and must pass a two-week shadow mode validation first. The system suppresses repeated detections within a defined time window, filters events below confidence thresholds, and routes alerts to the right people rather than broadcasting to everyone. The goal is that every alert that reaches someone means something.

They can. Protex can ingest external data — shift rosters, permit status, machine state from PLCs — and use it to determine whether a detection is actually actionable. A zone that's restricted during production can automatically open during a scheduled maintenance window, for instance.

Forklifts (reach trucks, counterbalance), manual and electric pallet jacks, AGVs, AMRs, HGVs, delivery vans, and trailers. Pedestrian movement is monitored alongside vehicles to analyse human-machine interactions and flow patterns. Each object gets a unique tracking ID maintained across the camera view, even at the distances typical in loading docks or yard environments.

No retraining required. Models come pre-trained on industrial environments and adapt to your site through configuration — you adjust zones, thresholds, dwell times, and object sensitivity through visual interfaces. During implementation there's a dedicated calibration period where the system refines to your specific lighting, camera angles, and layout. After that, model updates deploy daily automatically.

The platform tracks precision, recall, false positive rate, and a relevancy metric (which tends to matter more in practice than raw accuracy). When customers flag false positives, those go directly into the model refinement pipeline. Improvements are typically deployed within hours to days. Over time, false positive rates drop as the system learns what's operationally relevant at your site versus what's technically a detection but not actually a problem.

No. All video is processed on the on-site edge device. Only short encrypted event clips and metadata — a few seconds of footage around a detection — are sent to the cloud platform. Raw video stays on your network.

ISO/IEC 27001:2022 (independently audited annually), GDPR, UK DPA, and CCPA compliance, and alignment with the EU AI Act. The platform also holds a CSA STAR Level 1 self-assessment. Security in transit uses TLS 1.2 or higher; data at rest is encrypted with AES-256

Four options: facial blurring (face only, body visible), full-body blurring, silhouette ghosting (person replaced with a neutral outline), and videoless mode (video is processed at the edge for detection but never recorded or transmitted at all). Access to any footage is restricted to authorised personnel and logged via role-based access control with MFA.

Customisable dashboards at the camera, shift, and event-type level; heatmaps showing where risk or congestion concentrates within a facility; trend reports tracking events and operational patterns over time; and automated notifications triggered by specific event thresholds. Protex Intelligence lets teams query site data in natural language and export outputs in PPT, PDF, or custom formats.