Sentinel Multi‑Cam AI for edge-first security teams.
Ingest up to eight live camera feeds, recognise people, faces, objects, and vehicles in real time, and orchestrate explainable alerts with auditable history. Sentinel keeps sensitive telemetry on-site while empowering operators with actionable intelligence.
Operational clarity at the edge.
Sentinel Multi‑Cam AI is a desktop and edge appliance experience that prioritises privacy, resilience, and operator trust. Every detection is paired with context, confidence, and the ability to interrogate the full event timeline.
Core capabilities
- Face recognition with instant grouping of unknown visitors
- Shirt and object colour sampling to accelerate re-identification
- Per-object state machine: placed → picked up → held → returned
- Vehicle registry with perceptual hashing, metadata, and snapshots
- Review queues for people, faces, vehicles, and sensitive events
Architecture at a glance
- Inference: OpenCV + YOLOv3 objects, face_recognition (dlib) for faces
- Runtime: Tkinter desktop UI with multi-threaded workers per camera
- Acceleration: CUDA when available, optimised CPU fallbacks when not
- Storage: Local JSON/NPY & snapshots; optional managed DB connectors
- Extensibility: Webhooks, custom labels, colour pooling per label
Feature map
Every module is engineered to provide visual context, configurable sensitivity, and explainable actions so your security teams can focus on response rather than guesswork.
People & Faces
Identify known faces, consistently bucket unknowns, and pair each detection with a descriptive shirt colour for rapid escalation.
- Known/unknown persons with last-seen recall
- Hex + named colour sampling near detected heads
- Cooldowns & spot-locks prevent alert fatigue
Objects & Actions
Each object maintains its own state machine, logging precise pick-up, hold, and return events with durations and custom labels.
- Custom labels for objects or object+colour pairs
- Auto-place timeout when items leave frame
- Spot-locks to suppress duplicate re-tags
Vehicle Registry
Maintain a local, privacy-respecting log of vehicles with snapshots, perceptual hashes, and operator annotations.
- Colour detection & hex values
- Recent sighting timeline with export
- Snapshot viewer for verification
Alerting & Notifications
Deliver the right notification to the right channel with configurable cooldowns and optional built-in media hosting.
- Pushover, Telegram, Twilio SMS/MMS & Webhooks
- Subscribe by person, object, vehicle, or label
- Global cooldowns to prevent alert storms
Human-centric AI with explainability at the core.
Every detection can be traced, tuned, and justified. Sentinel tracks actions, context, and historical fingerprints so operators can trust automated interventions.
Multi-modal recognition
Fused computer vision for every asset in motion.
- Face recognition with unknown re-identification buckets
- People tracking with shirt colour sampling near heads
- YOLO-based object and package recognition with custom labels
Events that read like plain English
Log what actually happened — not just raw frames.
- Picked-up, held, and placed events with precise durations
- Spot-locks to suppress repetitive tagging in stable scenes
- Auto-place timeouts to resolve lingering items
Alerting that reaches your team
Turn detections into responses instantly.
- Webhook integrations for SOC dashboards
- Push notifications via Pushover, Telegram, and Twilio SMS
- Local queues for faces, people, and vehicles with quick actions
Forward roadmap
- Q4ONVIF discovery & RTSP presets for rapid camera onboarding
- Q1Optional cloud sync (events only) with granular role-based access
- Q1Pluggable model upgrades (YOLOv5/8, RT-DETR) with auto-benchmarks
- Q2Fine-grained privacy policies per camera and selective redaction
Built for on-prem assurance.
Sentinel runs at the edge with hardware-aware optimisations. Whether you deploy on CUDA GPUs or Intel CPUs with OpenVINO, the stack is tuned for sub-40 ms responsiveness.
Edge stack
- Core: OpenCV pipelines with YOLOv3 for object detection and
face_recognition(dlib) for faces - App: Multi-threaded desktop client with Tkinter UI and GPU-aware workers
- Database: Local SQLite event store with perceptual hashing for vehicle snapshots
- Explainability: Threshold tuning overlay, heat maps, idle percentage controls, and NMS tuning
Deployment blueprint
- Runs on Windows or Linux workstations with NVIDIA CUDA or Intel iGPU acceleration
- Supports RTSP, USB, and file-based camera ingest with adaptive frame rates
- Edge compute hardened for air-gapped networks — no outbound traffic required
- Optional integration with Ioncore energy storage for redundant runtime
Plans built for every deployment stage.
Licences are per host with up to eight cameras per node. Keep all telemetry on-site, or connect your own cloud datastore for retention analytics.
Starter
Perfect for pilot programmes and boutique spaces.
- Up to 3 cameras per host
- Local storage & export tools
- Email support
Pro Most Popular
Full telemetry with premium notifications and media hosting.
- Up to 8 cameras per host
- All notification channels + MMS host
- Priority email support
Enterprise
Tailored orchestration, governance, and service levels.
- Multi-host deployments & orchestration
- SAML/SSO, audit exports, hardened ops
- Dedicated success & SLA coverage
Prices exclude optional cloud databases or object storage. Use the estimator below to model your retention and egress strategy.
Deployment licences aligned with rollout velocity.
Scale from single-site pilots to enterprise-wide coverage with support, training, and governance baked in.
Ideal for a single facility or proof-of-concept.
- Up to 4 camera streams
- Standard alerting channels
- Quarterly model refresh
- Email support
Designed for multi-site teams that need resilience.
- Up to 8 camera streams
- Custom object libraries
- 24/7 alerting with failover
- Phone & chat support
Tailored to high-security installations with compliance needs.
- Unlimited camera clusters
- Dedicated success engineer
- Private model training
- On-site commissioning
Database & storage cost estimator.
Forecast retention costs for event logs and snapshots. Replace the assumptions with your provider's official pricing.
Event logs
Snapshots
Rates
Base fees
Illustrative only. Adjust for your provider and compliance needs.
Estimate your event storage footprint.
Model how Sentinel compresses data across locations while maintaining a complete audit trail.
Ioncore field teams benchmarked the Operations tier at roughly 3.9 GB for 30 days of retention across 8 cameras with snapshot attachments. Your actual footprint adjusts linearly with event density and resolution.
Frequently asked.
Answers to the most common Sentinel deployment questions.
Do I need a cloud database?
No. Sentinel is local-first. Cloud storage is optional for centralising analytics or unifying multi-site telemetry.
How do notifications work?
Subscribe to people, objects, or vehicles and route alerts via Pushover, Telegram, Twilio SMS/MMS, or Webhooks. Built-in cooldowns keep noise low.
What about privacy and compliance?
You control retention and export. Logs are auditable, spot-locks limit repeat tagging, and future privacy profiles enable per-camera policies.
Can I change detection models?
Yes. Today’s default stack is YOLOv3 for objects and face_recognition for faces. The roadmap introduces pluggable backends with automatic benchmarking.
Can Sentinel run completely offline?
Yes. Sentinel is engineered for air-gapped deployments. All inference, storage, and review workflows run locally. Optional alerting channels can be routed through on-prem messaging servers.
How does it handle unknown individuals?
Unknown faces are bucketed with re-identification embeddings, letting analysts group recurring visitors and promote them to known profiles when appropriate.
What about cybersecurity?
Sentinel is delivered as a hardened desktop package with signed binaries. Audit logs capture administrative actions, model updates, and integration changes for compliance teams.
