Your operators watch 200 screens.
They see 10.
AICU is the brain your control room is missing. One AI system that watches every camera, understands what it sees, and tells your operators exactly what to do — in under 5 seconds.
Unknown male tracked across 3 cameras over 6 minutes.
Not in resident or staff database.
Same person who tested your fence 3 weeks ago. (Phantom Memory — 87% match)
Detection to dispatch: 30 seconds.
Your control room watches 200+ cameras. After 20 minutes, operators miss 90% of events.
Your clients pay for proactive security. They get reactive investigation.
AICU closes the gap. One system, every camera, every second.
The economics of human-only monitoring are broken. A well-trained operator can sustain focused attention on a maximum of 12-16 camera feeds simultaneously. After 20 minutes of continuous monitoring, cognitive fatigue sets in and detection rates collapse. Studies consistently show operators miss over 90% of security events after the first 20 minutes of a shift.
Hiring more operators does not solve the problem — it scales the cost without scaling the attention. At 200 cameras, you need 12-16 operators just to cover every feed with one pair of eyes. At 500 cameras, it becomes economically impossible.
The reality is that most security companies are running a recording service, not a monitoring service. Cameras capture footage. Nobody watches it until after something happens. The client pays for proactive protection and receives a post-incident evidence retrieval system.
AICU changes this equation. One AI system watches every camera, every second, with zero fatigue. Your operators shift from watching screens to acting on intelligence.
Phantom Memory
The guy from last month is back. Instant recognition.
Ghost Alert
Expected at camera 7. Never arrived. Where is he?
Cross-Camera Tracking
One person. Every camera. No face needed.
Intent Detection
Casing walk. Climbing prep. Body language doesn't lie.
Load Shedding Intelligence
Power's out. Everyone's blind. Except you.
Threat Escalation
Perimeter → chokepoint → house. Score rising. CRITICAL.
Real-time alerts fire instantly. The Unknown Persons Gallery is our quiet intelligence layer — it only shows candidates after 5+ quality observations.
Every other system tells you what happened. AICU tells you what's about to happen.
Detection is table stakes. Escalation is better. But Foresight is what no one else has. AICU continuously analyses what it sees, what it expected to see, and what it remembers — then alerts your operators to threats that haven't fully materialised yet.
AICU knows your camera layout. It knows which cameras a person should appear on next, and how long it should take them to get there. When someone is detected heading toward a chokepoint — and they never arrive — AICU fires a Ghost Alert.
They didn't disappear. They stopped. They're hiding in the gap between your cameras — and AICU is the only system that knows they should have been seen but weren't.
Three weeks ago, an unidentified male was seen testing your client's perimeter at 2am. The alert was dismissed. His physical signature — body shape, height, build, walking pattern — was archived. Tonight, someone matching that signature appears again.
Your operator isn't starting from zero. They have context from a previous incident they may not even remember. AICU remembers everything.
AICU reads body language through skeletal analysis and flags behaviour patterns that indicate hostile intent. These are deterministic rules applied to measured body positions — not guesses.
| Behaviour | AICU Flags | Meaning |
|---|---|---|
| Repeated direction changes + head scanning | Casing behaviour | Studying property layout and camera positions |
| Crouch near wall + arms reaching upward | Climbing preparation | About to scale a wall or fence |
| Low posture + slow movement near perimeter | Stealth approach | Attempting to avoid detection |
| Asymmetric posture + slow gait | Carrying heavy object | Possibly removing stolen property |
| Direct path then sudden retreat | Testing response | Checking if anyone reacts to their presence |
AICU knows where every person is, how fast they're moving, and which direction they're heading. Combined with the property layout, it predicts where they'll be in 30, 60, 90 seconds.
Your patrol doesn't chase where the person was. They intercept where the person is going.
After two weeks, AICU learns the rhythm of every client site. When reality deviates from the baseline, it flags it — even if no individual detection would trigger an alert on its own.
Three people at a camera that usually sees nobody at that hour? That's not just a detection. That's a pattern. That's intelligence.
During a power outage, most security systems go dark. AICU doesn't. But more importantly — AICU knows that movement during load shedding is inherently more suspicious than movement during normal conditions.
While the suburb was dark and every other system was offline, AICU was watching — and it knew that movement during an outage means something different.
| Capability | Traditional AI | AICU Foresight |
|---|---|---|
| Person expected but missing (blind spot) | ❌ | ✅ Ghost Alert |
| Person was here weeks ago | ❌ | ✅ Phantom Memory |
| Hostile body language detected | ❌ | ✅ Intent Detection |
| Predicts where person is heading | ❌ | ✅ Trajectory Prediction |
| Abnormal activity for time/place | ❌ Most | ✅ Anomaly Detection |
| Movement during power outage flagged | ❌ | ✅ Load Shedding Intelligence |
| 30 sites at a glance | ❌ | ✅ Ambient Threat Level |
AICU does not just detect people. It tracks their trail across your client site and escalates the threat score as they move deeper into secured zones. Every escalation is deterministic. Rules, not models.
| Movement Pattern | Zone | Threat Score | AICU Response |
|---|---|---|---|
| Person detected on perimeter | Perimeter | Low (0.2) | Log event, start tracking |
| Same person moves to internal road | Transition | Medium (0.4) | Flag for operator attention |
| Enters chokepoint (gate area, pathway) | Chokepoint | High (0.65) | Active alert, recommend observation |
| Approaches building / restricted zone | Building Access | Critical (0.85) | Immediate dispatch recommendation |
| Loitering near door/window > 30s | Building Access | Critical (0.95) | Urgent alert, historical context shown |
| Match to prior incident subject | Any | +0.3 modifier | Escalate immediately with incident history |
30 sites at a glance. One colour per site. Your operator knows where to focus.
"Every other system tells you what happened. AICU tells you what's about to happen."
AI is probabilistic. It observes and scores. It's allowed to be wrong.
Rules are deterministic. Written by your team. Fully auditable.
Humans have final authority. Always.
AICU runs on AIOS — the AI Operating System. Every decision flows through three distinct layers, each with a different job and a different accountability model.
AI Perceives
Neural networks detect, classify, and track every person, vehicle, and object across every camera feed. Outputs are confidence scores, not decisions. The AI is allowed to be wrong — that is what probability means.
Rules Decide
Human-defined rules evaluate AI perceptions against thresholds, zones, schedules, and policies. Not learned. Not probabilistic. Not opaque. Every rule is written by your team, auditable, and predictable.
Humans Act
Operators receive intelligence briefings — not raw alarms. They see the trail, the description, the context, the confidence, and the recommended action. They confirm, dismiss, or escalate. Humans always have final authority.
AI is allowed to be wrong. Rules are always predictable. Humans always have final authority.
This architecture satisfies compliance officers, insurance underwriters, and lawyers. When your operator asks "why did this alert fire?" — the answer is a traceable chain: the AI perceived X with Y confidence, rule Z evaluated to true, the briefing was delivered. Full transparency. Every time.
Two boxes in your server room. Each with a distinct role. Connected to each other, isolated from everything else.
Detection runs on low-resolution sub-streams (640×360) for efficiency — 12 cameras at sub-stream resolution is trivial GPU load. When something is detected, AICU pulls a full-resolution frame from the main stream for deep analysis. This sub-stream/main-stream split is how 200+ cameras become feasible on modest hardware.
Ingestion Node
Sits on the camera VLAN. Lightweight GPU. Its only job: pull RTSP sub-streams from every camera, run fast detection at 30-50 FPS, and forward detections to the analysis node.
Pulls sub-streams (low resolution, low bandwidth) for continuous monitoring. When a detection occurs, it requests the main-stream clip (full resolution) for that camera and forwards it for deep analysis.
Analysis Node
Server rack mount. Powerful GPU. Runs the full intelligence pipeline: re-identification, face recognition, plate OCR, skeleton analysis, gait matching, threat scoring, rule evaluation, and operator briefing generation.
Never touches the camera network directly. Receives only processed detections and full-resolution event clips from the ingestion node.
Security
The analysis node never touches the camera VLAN. Compromising one node does not expose the other.
Scalability
Add more ingestion nodes for more cameras without replacing the analysis node. Add analysis nodes for more intelligence.
Resilience
If the analysis node goes down, the ingestion node continues recording detections. When it recovers, it processes the backlog.
Performance
Detection and analysis are different workloads. Splitting them across dedicated GPUs means neither bottlenecks the other.
If the analysis node goes down, the ingestion node keeps detecting and recording. If a face recognition model crashes, detection and tracking continue uninterrupted. No single component failure takes the system offline. Your cameras are always being watched.
AICU does not rely on a single signal. It extracts multiple independent signatures from every person and vehicle, fusing them into a composite identity that works even when individual signals are weak.
8 out of 10 cameras can't see a face. AICU still tracks across all of them.
| Signal | What AICU Extracts | Min. Camera | Used For |
|---|---|---|---|
| Body fingerprint | 2048-dimension appearance embedding (proportions, colours, shape) | 720p | Cross-camera re-identification |
| Skeleton | 17-point pose estimation (joints, limbs, posture) | 720p | Height estimate, build classification, posture |
| Clothing | Upper/lower garment colour and type extraction | 720p | Physical description, re-identification |
| Gait | Step cadence, stride length, walking rhythm from keypoint sequence | 1080p | Behavioural signature, persistent identity |
| Face | 512-dimension face embedding when visible | 2MP+ close | Positive identification, database matching |
| Direction | Heading vector and speed from tracked path | 720p | Movement prediction, zone transition detection |
Height and distance estimates are calibrated per camera using physical ground-truth measurements during installation — not guesses from pixel ratios. When AICU reports '~182cm', that number is calibrated against real-world reference points at that camera's specific position and angle.
| Signal | What AICU Extracts | Min. Camera | Used For |
|---|---|---|---|
| Number plate | SA-optimised OCR, dirty/partial plate handling | 1080p close | Vehicle identification, database matching |
| Colour | Body colour extraction under variable lighting | 720p | Description, re-identification |
| Type | Car, bakkie, SUV, motorbike, truck classification | 720p | Filtering, context |
| Make / model | Brand and model estimation from shape signature | 1080p | Detailed identification |
| Appearance fingerprint | Whole-vehicle embedding (shape, colour, features) | 720p | Cross-camera vehicle tracking without plate |
When AICU sees a person on one camera and a person on another, it does not guess. It fuses every available signal into a single confidence score.
Matches are reported as CONFIRMED (90%+), PROBABLE (70-89%), POSSIBLE (50-69%), or WEAK (below 50%). Only PROBABLE and above generate operator alerts by default. Thresholds are adjustable per site.
One person. One trail. One identity.
No other system in South Africa does this. When a figure appears on a perimeter camera at 2am, AICU follows them across every camera on the client site — without ever seeing their face.
Traditional face recognition systems require manual enrollment — uploading photos of every resident, every staff member, every contractor before the system can recognise anyone. At 50 cameras, that's tedious. At 500 cameras with hundreds of unique people per day, it's impossible.
AICU inverts this entirely. It discovers people through normal operation and presents them to the operator as candidates ranked by importance. The AI does the work. The operator makes the decision.
| Observed Pattern | AI Suggests |
|---|---|
| Appears every weekday 7am-4pm at the same site | Staff |
| Appears once at 2am at the perimeter fence | Suspicious |
| Arrives in a registered resident vehicle | Visitor |
| Seen at 3 different managed sites in 2 weeks | Flagged — possible reconnaissance |
| Appears every Tuesday and Friday morning at the gate | Regular contractor |
| Seen only on weekends during daylight hours | Weekend visitor |
The dashboard presents unclassified candidates in a sortable gallery — ranked by frequency (most seen first), threat level (highest risk first), or recency (newest first). Candidates must have 3+ sightings or 1 high-quality face crop before appearing in the queue. Single-appearance passersby silently expire without ever bothering the operator.
At 1,000 cameras with hundreds of unique people per day, the operator's job isn't to enrol everyone. It's to classify the ones that matter. AICU handles the rest.
Entities follow a managed lifecycle — Active → Cooling → Stale → Archived → Resolved. Active tracking expires automatically when the person leaves. Archived data is purged per POPIA retention policies. No manual cleanup needed.
Every confirmed or dismissed alert teaches AICU about your specific cameras and conditions. Operator feedback — correct alert, false positive, wrong match, trail split — accumulates as labelled data. Confidence accuracy is checked daily; threshold calibration runs weekly using accumulated feedback. False alarm rates drop week over week. This is not AI retraining — it is systematic calibration driven by your operators' real-world decisions.
The AI discovers. The AI suggests. The operator decides. Same principle. Every layer.
Most AI systems produce alerts. AICU produces evidence.
Every decision is logged, every reasoning chain is preserved, every piece of evidence is cryptographically sealed.
When a lawyer asks "why did this alert fire?" — you show them the ledger. Full transparency. Every time.
Decision Ledger
Every alert, every score change, every classification — logged with full inputs, weights, and reasoning. Audit any decision after the fact.
Tamper-Proof Evidence
SHA-256 hash chain on all evidence. If a single frame is altered, the chain breaks. Legally defensible.
Forensic Freeze
One click preserves everything about an incident. Exempt from retention policies until released. Ready for SAPS, insurance, or court.
Every decision AICU makes is recorded in an append-only ledger. Not just what happened — but why. Every signal that fed the decision, every weight that was active, every modifier that applied, and every signal that was available but not used.
| Recorded For Every Decision | Why It Matters |
|---|---|
| All input signals + quality scores | Prove what the system saw and how clearly it saw it |
| Active weights and modifiers | Show which rules influenced the score and by how much |
| The decision output | The exact threat score, classification, or alert that resulted |
| Signals available but not used | Prove the system didn't ignore relevant evidence |
| System mode at the time | Was the system in normal, demo, safe, or degraded mode? |
| Human override (if any) | Which operator overrode the system, and when |
When your client's insurance asks why patrol was dispatched, you don't explain — you show them the ledger entry. Inputs, weights, output, rule chain. Case closed.
Every piece of evidence — every detection frame, every alert, every operator action — is linked in a SHA-256 hash chain. Each entry includes the hash of the previous entry. If a single record is modified or deleted, the chain breaks and the tampering is immediately detectable.
When an incident matters — break-in, armed robbery, insurance claim — your operator hits one button. AICU freezes everything related to that incident: every entity snapshot, every modifier state, every decision ledger entry, every video clip, every alert. All of it exempt from automated retention policies until explicitly released.
POPIA audit log records every freeze and release. Your compliance team has a paper trail. Your client has evidence designed for legal defensibility.
Perimeter cameras continuously buffer the last 30 seconds of footage. When a detection fires, the buffer is saved alongside the event. You don't just see the moment of detection — you see how they got there. Where did they come from? Did they arrive on foot or from a vehicle? Were they alone?
The most important 30 seconds are the ones before the system noticed. AICU captures them automatically. During load shedding, the buffer switches to priority cameras only to conserve UPS power.
"We can explain every alert after the fact. Can your current provider?"
No rip and replace. Dahua, Hikvision, or whatever you run today. If it streams video, AICU processes it.
| Camera Type | Connection | Detection | Tracking | Face / Plate ID |
|---|---|---|---|---|
| Analog (via DVR) | RTSP from DVR | Yes | Yes | No |
| 720p IP | Direct RTSP | Yes | Yes | Limited |
| 1080p IP | Direct RTSP | Yes | Yes | Yes (close range) |
| 2MP+ IP | Direct RTSP | Yes | Yes | Yes |
| 4MP+ IP (gate/entrance) | Direct RTSP | Yes | Yes | Yes (full range) |
| PTZ | Direct RTSP + ONVIF | Yes | Yes + auto-follow | Yes |
| Thermal | RTSP | Yes | Yes | No (body only) |
Perimeter
Detection, tracking, body fingerprint, direction. Full trail intelligence even at low resolution.
Chokepoint
Everything above plus gait analysis, better clothing detail, improved re-identification confidence.
Gate / Entrance
Full intelligence: face recognition, plate OCR, vehicle make/model, positive identification.
Tell your clients: you do not need to replace a single camera.
Your old system records everything and finds nothing. AICU records continuously at low resolution for compliance, and triggers full-resolution event clips only when something happens.
| Traditional NVR | AICU | |
|---|---|---|
| Recording mode | 24/7 full resolution | Continuous low-res + AI-triggered full-res events |
| Storage per camera/day | ~40-60 GB (4MP H.265) | ~8-12 GB combined |
| 30-day retention (100 cameras) | ~150-180 TB | ~30-35 TB |
| Finding an incident | Manual scrubbing, hours | AI-indexed, instant search by description |
| Evidence quality | Full resolution but hard to find | Full resolution, pre-tagged, exportable |
| Storage cost (100 cameras, 30 days) | R120K+ in drives | R25K in drives |
AICU handles both detection and recording. For most deployments, your existing NVR becomes redundant — AICU replaces it entirely with smarter, searchable, AI-annotated storage at a fraction of the disk usage.
AICU measures its own accuracy. If it's been saying "confirmed" but getting it wrong — it automatically raises the bar and tells your operator.
The most dangerous thing an AI security system can do is sound certain when it's wrong. AICU makes that impossible.
Self-Calibrating Confidence
Compares its predictions to outcomes. If accuracy drifts, it auto-corrects and tells the operator.
Anti-Evasion Intelligence
Clothing changes, face covering, route avoidance — evasion attempts become the signal.
Commercial Proof Layer
Monthly value report: hours saved, false alerts prevented, unknowns surfaced. Numbers that close renewals.
Compound Intelligence Moat
Every day of data makes AICU harder to compete with. A competitor starting today can't copy 12 months of accumulated site intelligence.
Every 24 hours, AICU compares its confidence predictions to actual outcomes. If 15% of "CONFIRMED" calls turned out to be wrong matches, the system automatically tightens the threshold and tells the operator: "I've been less accurate than I should be. I'm being more conservative until I recalibrate."
| Confidence Drift | AICU Response |
|---|---|
| < 5% error rate | Healthy. No action needed. |
| 5-10% error rate | Logged in weekly digest. Flagged for review. |
| 10-20% error rate | Auto-shifts all confidence bands down one level. Narrates: 'I'm being more conservative.' |
| > 20% error rate | Triggers safe mode. Narrates: 'My confidence is unreliable. Safe mode enabled.' |
Most AI systems are confidently wrong and never know it. AICU literally cannot become overconfident — because it checks itself every day.
Every other security AI gets dumber when someone actively tries to fool it. AICU gets more suspicious. The attempt to evade IS the intelligence.
| Evasion Technique | AICU Detects | Response |
|---|---|---|
| Changed clothing between cameras | Same skeleton + gait, different colour histogram | "Clothing changed. Body match 84%. Suspicious." |
| Face covering (mask, hoodie) | Face expected at close camera but not captured | "Face should be visible but wasn't. Possible concealment." |
| Altered walking pattern | Gait variance spikes but skeleton proportions match | "Gait inconsistency. Same build. Possible deliberate evasion." |
| Took the blind-spot route | Entity trail skips cameras it should have passed | "Went from 206 to 210 without 207. Took the blind spot path." |
Operator feedback makes AICU smarter — but only when that feedback is reliable. AICU tracks each operator's accuracy over time. If one night-shift operator marks everything as "false positive," the system reduces their influence automatically and flags it for admin review.
| Operator Trust | System Behaviour |
|---|---|
| Trusted (> 85%) | Full weight on overrides. Auto-apply classification changes. |
| Standard (60-85%) | Normal weight. Changes apply but flagged for review. |
| Uncertain (40-60%) | Reduced weight. Changes require admin confirmation. |
| Unreliable (< 40%) | Changes logged only. Admin notified for review. |
Every month, AICU generates a value report for each client site. This is not marketing — it is measured, verifiable data from actual operations. Send it to your client. Let the numbers speak.
AICU shows its own maturity in the dashboard header. This makes the compound intelligence advantage visible — to operators and to buyers.
| Intelligence Age | System Maturity |
|---|---|
| 0-6 days | Learning — building initial baselines |
| 7-13 days | Baselines forming — anomaly detection activating |
| 14-29 days | Anomaly detection active — patterns emerging |
| 30-89 days | Classifications reliable — system knows who belongs |
| 90-179 days | Seasonal patterns emerging — winter vs summer baselines |
| 180+ days | Full intelligence — comprehensive historical knowledge |
A competitor who starts today arrives at week one when you've been running for a year. They can copy the software. They cannot copy twelve months of accumulated intelligence about your properties.
"AICU may become smarter over time, but it may never become less explainable."
Load Shedding Resilient
Monitors UPS battery, adjusts processing priority, safe shutdown at 30%, automatic recovery within 90 seconds when power returns. Your clients stay protected through every stage.
SA Camera Ecosystem
Optimised for Hikvision, Dahua, and SA-dominant DVR/NVR systems. If it outputs RTSP, it works. No proprietary lock-in.
SA Number Plates
Plate recognition tuned for South African formats, dirty plates, non-standard mounting, and variable lighting conditions.
POPIA Compliant
Fully on-premise. Biometrics encrypted at rest (AES-256). Automated retention policies. Consent tracking. Full audit logs. Data subject access request support.
Local Support
KwaZulu-Natal based. South African owned. On-site support, not an international ticket system 8 timezones away.
Priced in Rands
No dollars. No exchange rate surprises. No international billing complications. Your margins are predictable.
| UPS Level | AICU Action |
|---|---|
| 100-60% | Full operation — all cameras, all intelligence |
| 60-40% | Reduced non-critical processing, priority cameras maintained |
| 40-30% | Priority cameras only, minimal processing |
| 30% | Safe shutdown with state saved |
| Power returns | Automatic recovery within 90 seconds |
| Requirement | AICU Implementation |
|---|---|
| Data sovereignty | Fully on-premise — no data leaves South Africa |
| Biometric protection | Encrypted at rest (AES-256) |
| Retention limits | Automated retention policies, configurable per site |
| Consent tracking | Consent management for face/plate enrollment |
| Audit trail | Full audit logs for every access and decision |
| Data subject access | Support for data subject access requests |
| Purpose limitation | Restricted to security operations only |
Every site is different. AICU adapts to your client's property, camera layout, and security profile — not the other way around.
And when something breaks, the system gets more suspicious — not less. Every failure mode errs on the side of alerting.
Site Acceptance Scorecard
Pre-deployment assessment: camera quality, coverage gaps, environment score. Your client knows what to expect.
Policy Packs
One config for estates, another for commercial yards, another for armed response control rooms. Platform, not product.
Day One Intelligence
New sites don't start from zero. Learned priors from similar properties mean useful intelligence from day one.
Fail-Closed Security
When a component fails, threat scores go up — not down. A broken system is more suspicious, not less.
Before every deployment, AICU assesses the site and produces a readiness scorecard. Camera quality per position, face recognition availability, coverage gaps with exact distances, environment factors, and power resilience. Your client gets an honest, professional assessment — not a sales pitch.
"Your site scores 78/100. Face recognition works at these 4 positions. Plates at these 2. Body tracking everywhere. Here's what to fix to get to 90+." Professional, consultative, honest.
A residential estate with quiet-hour rules is not a 24/7 commercial yard. AICU ships with pre-built policy configurations for different security environments. Select a pack, adjust to your client's needs, deploy.
When AICU deploys to a new site, it transfers structural knowledge from existing deployments. Not biometric data — never. But patterns that are universal: "Mondays are quieter than Fridays," "delivery drivers stay 5-15 minutes," "load shedding stage 4 means +20% activity."
Instead of 7 days to form baselines, the system has useful priors on day 1. Baselines converge 2-3x faster because they start from informed priors, not from scratch. POPIA-safe — no biometric data ever leaves a site.
When a component fails, AICU does not silently degrade. It escalates. If the system can't identify someone, it treats everyone as unknown. If intelligence is offline, classification-based suppressions are disabled. No one gets a free pass because the system is having a bad day.
| Component Failure | AICU Response |
|---|---|
| Signal extraction down (no face, body, plate) | All threat scores raised 1.3x — can't identify anyone, treat all as unknown |
| Intelligence engine offline | All classification suppressions disabled — no entity gets a free pass |
| Matching uncertain (low confidence) | Escalate to operator — never auto-dismiss an uncertain match |
| Camera offline | Alert operator, auto-reconnect attempts, gap logged in coverage report |
| System health critical (< 30) | Auto-enables safe mode, freezes all intelligence evolution, operator notified |
| Power outage detected | UPS-aware shutdown sequence, auto-recovery within 90 seconds |
Every failure mode has been designed. Nothing is left to chance. Your operators always know the system's state, and the system always errs on the side of safety.
"The difference between a demo that impresses and a system that survives production."
On-premise. Your data never leaves your building.
Trail-based threat scoring — not just detection, but escalation across zones.
Works with your existing cameras. No rip and replace.
Full decision transparency. Every alert has a traceable rule chain.
| Feature | Smart Camera | Cloud AI | AICU |
|---|---|---|---|
| Works with existing cameras | No — buy their cameras | Some — limited brands | Yes — any RTSP camera |
| Cross-camera tracking | No | Limited (face only) | Yes — body, gait, clothing, face |
| Data stays on-premise | Varies | No — cloud processing | Yes — always |
| Decision transparency | Black box | Black box | Full rule chain audit |
| Load shedding resilient | No | No (needs internet) | Yes — UPS-aware, auto-recovery |
| SA plate recognition | Basic | Poor (US/EU trained) | Yes — SA-optimised |
| Trail-based threat scoring | No | No | Yes — perimeter to building |
| Multi-entity coordination | No | No | Yes — group & vehicle linking |
| POPIA compliant | Varies | No (data leaves SA) | Yes — full compliance |
| Internet required | No | Yes — always | No — fully offline |
| Pricing model | $$$ (USD/EUR) | $$$/month (USD/EUR) | Rands. Transparent. Contact us. |
| Local support | Distributor | International tickets | On-site, KZN-based |
Real threats rarely involve one person. AICU detects coordinated behaviour across multiple individuals and vehicles — patterns your operators would never see across dozens of camera feeds.
| Pattern | What AICU Detects | Significance |
|---|---|---|
| Coordinated approach | 2+ subjects entering from different directions within time window | Planned entry — not coincidence |
| Getaway pattern | Vehicle stationary with engine running + person on foot nearby | Staged escape vehicle |
| Group arrival | Multiple individuals arriving together, then splitting to cover different areas | Pre-planned reconnaissance or coordinated action |
| Cross-site vehicle | Same vehicle plate or appearance seen at multiple client sites | Multi-site target assessment |
| Unknown with resident | Unknown person arrives with known resident — threat score reduced | Contextual de-escalation, fewer false positives |
| Approach-retreat | Subject approaches secured zone, reverses direction, approaches again | Testing response times, probing defences |
AICU integrates directly with armed response and control room platforms via standard webhooks. When your operator confirms a dispatch, the alert is sent automatically — no re-typing, no phone calls, no delays.
Compatible with Listener, Patriot, Mastermind, and any platform that accepts HTTPS webhook notifications.
{
"event": "dispatch_request",
"priority": "critical",
"site": "Oceanview Estate",
"zone": "Unit 14 — North Access",
"threat_score": 0.88,
"subject": {
"height_cm": 182,
"build": "medium",
"upper_clothing": "dark blue",
"lower_clothing": "grey",
"carried": ["backpack"],
"direction": "SE",
"tracked_minutes": 6,
"cross_camera_confidence": 0.91
},
"trail": [
{"time": "02:10", "camera": "Cam 7", "zone": "perimeter"},
{"time": "02:12", "camera": "Cam 3", "zone": "internal_road"},
{"time": "02:14", "camera": "Cam 12", "zone": "chokepoint"},
{"time": "02:16", "camera": "Cam 9", "zone": "building_access"}
],
"rule_chain": "unknown_person + building_zone + after_hours + score>0.85",
"historical_matches": 2,
"recommended_action": "Dispatch patrol to Unit 14 north side"
}Without AICU, one operator effectively monitors 10-20 cameras. With AICU, one operator manages 200+ — because the AI handles perception and the operator handles decision-making. You don't need more operators. You need smarter tools.
Offer AI-enhanced monitoring as a premium tier. Your margins grow with every client that upgrades. AICU is not a cost — it is the foundation of a service your competitors cannot match.
All pricing in South African Rands. No exchange rate surprises. Final pricing depends on site assessment.
Ready to see what your cameras are missing?
[email protected] · KwaZulu-Natal, South Africa
No. AICU works with whatever cameras your clients already have. If they stream RTSP (90%+ do), they work. Analog cameras on a DVR also work — the DVR converts to RTSP. Tell your clients: you do not need to replace a single camera.
No. AICU runs entirely on-premise. Both nodes sit in your server room. No cloud processing. No data uploaded anywhere. Your clients' footage never leaves your network. This is a fundamental architectural decision, not a configuration option.
AICU does not make decisions. AI perceives (probabilistic, confidence scores). Human-defined rules decide what matters (deterministic, auditable). Your operators act on the intelligence briefing (final authority). Every alert has a traceable rule chain explaining exactly why it fired.
Fewer than you think. One operator with AICU can effectively manage 200+ cameras because the AI handles perception and the operator handles decision-making. AICU does not replace operators — it makes each operator dramatically more effective.
AICU monitors UPS battery levels and automatically adjusts operation. At 60% it reduces non-critical processing. At 40% it switches to priority cameras only. At 30% it performs a safe shutdown with state saved. When power returns, it recovers automatically within 90 seconds.
Nothing changes. AICU runs entirely on your local network. Internet is not required for any detection, tracking, or alerting functions. Internet is only used for optional remote dashboard access and software updates.
Yes. AICU sends alerts via webhooks (JSON/HTTPS) compatible with Listener, Patriot, Mastermind, or any system that accepts HTTP notifications. Priority levels, alert routing, and escalation rules are all configurable.
Yes. Each client site runs its own AICU hardware for local processing. A central dashboard in your control room provides cross-site visibility, unified threat intelligence, and consolidated reporting across all your clients.
AICU uses trail-based escalation specifically to reduce false positives. A single detection on one camera is logged but not alerted. Alerts only fire when behaviour escalates across zones. Operator feedback further tunes thresholds per camera and time of day. False positive rates drop steadily over the first weeks of operation.
50-200 cameras per GPU node with full intelligence (detection, tracking, re-identification, face, plates). Detection-only mode supports significantly more. Multiple nodes can be deployed for larger sites or multi-site operations.
We do. On-site installation, camera configuration, zone mapping, rule setup, and full operator training are all included. KwaZulu-Natal based — not an overseas ticket queue.
AICU is designed for full POPIA compliance: fully on-premise (no data leaves SA), biometric data encrypted at rest (AES-256), automated retention policies, consent tracking for face/plate enrollment, full audit logs, data subject access request support, and purpose limitation to security only.
No. AICU uses multiple independent signals — body proportions, skeleton, gait, height — not just clothing and face. If someone changes clothes between cameras, the body match still holds. If they cover their face, the absence of face data where it should be visible becomes a signal. Evasion attempts elevate the threat score, not reduce it.
Every alert is recorded in the Decision Ledger with full inputs, weights, modifier state, and reasoning chain. Evidence is linked in a SHA-256 cryptographic hash chain — tamper-proof and verifiable. If an insurance company or court asks 'why did this alert fire?' you show them the ledger entry.
AICU measures its own accuracy daily. When confidence predictions drift from actual outcomes, it automatically tightens its thresholds and tells the operator. The system literally cannot become overconfident because it self-corrects. Operator feedback further refines accuracy over time.
Yes. AICU ships with policy packs — pre-configured profiles for residential estates, private homes, commercial yards, and armed response control rooms. Each pack adjusts thresholds, quiet hours, escalation paths, occupancy models, and alert priorities. Select a pack, customise to the client, deploy.
Immediately. AICU transfers universal patterns from existing deployments — day-of-week activity, vehicle distributions, weather correlations — while keeping all biometric data site-local. Baselines form 2-3x faster than starting from scratch. Full anomaly detection typically activates within 2 weeks.
Yes. Every month AICU generates a Commercial Proof report per site: detections processed, false alerts suppressed, staff auto-recognised, unknown repeats surfaced, operator hours saved. Real numbers from real operations — not estimates. Send it directly to your client.