AI security that never leaves the building.
OricoAI turns any RTSP or ONVIF camera into an intelligent security system — detection, recognition, rules, alerts and an AI assistant — running entirely on your own hardware.
- Client
- Enterprise security operators
- Sector
- Physical security · Critical infrastructure
- Role
- Product engineering · AI · Platform
- Stack
- Vue 3 · Python · YOLO · ONNX / TensorRT / CoreML
Smart cameras got smart at your expense.
Modern camera “intelligence” lives in someone else’s cloud. Your footage leaves the building, the features hide behind a subscription, and your alarms depend on someone else’s uptime.
Privacy you can’t audit
Faces, plates and routines streamed to third-party servers with retention you don’t control.
Per-camera subscriptions
AI detection, person alerts and clip history locked behind recurring monthly fees, per device.
Latency & outages
A round-trip to the cloud before anyone reacts — and nothing works when the link is down.
Vendor lock-in
Closed ecosystems and cameras that stop being smart the day the service does.
One console. Any camera. Your hardware.
OricoAI is a complete, self-hosted AI video security platform — a normalization gateway, a rules engine, on-device detection & recognition, and a polished operator console. Install it on the box in your building and you own the whole pipeline.
Private by design
All inference runs locally. Footage and identities never leave your network — no cloud account.
Works with what you have
RTSP, ONVIF, webcams — any standard camera. No proprietary hardware, no rip-and-replace.
Hardware-accelerated
On-device YOLO, face & plate models run on your GPU / NPU via CoreML, TensorRT or ONNX.
Agentic, not rule-bound
Ask questions in plain language. An AI assistant explains activity, grounded in live events.
A clean stack, top to bottom.
Cameras feed a gateway that collapses every detection source — on-device models, MQTT, REST — into one canonical event stream. A rules engine evaluates that stream and pushes alerts to the console and your phone in real time. Everything runs in your network.
Nine screens built for an operations room.
A polished Vue 3 console that puts an AI security operator behind every camera — live, searchable, and answerable in plain language.
The whole site, at a glance.
A live threat level, the metrics that matter, and host telemetry — refreshed in real time over WebSocket.
- KPIs & 24-hour activity — Cameras online, active rules, events/hour, alerts and incidents — with a live activity chart.
- Detection mix & system load — Object / face / plate / fall breakdown beside live CPU, RAM and GPU utilization.
- Top firing rules — See which rules drive the most alerts and tune them without leaving the page.
Four AI engines, every frame, locally.
Every camera is analyzed on your own machine and normalized into one searchable event stream — no frame ever leaves the building.
- Searchable event stream — Filter by camera, type, label, confidence and time across the whole site.
- Object, face, plate & fall — Four engines run per frame and land in one canonical timeline.
- Jump to footage — Every event links straight to the clip that produced it.
Rules in your security team’s language.
Build rules on the events themselves — camera, type, label, confidence and time — then dry-run them against recent activity before going live.
- Structured rules — “Unknown person at night”, “missing employee badge”, “vehicle in zone” — with severity and actions.
- Smart rules (VLM) — Describe a situation in plain English — “a package is left at the door” — matched by a vision model against live scenes.
- Test before you trust — Dry-run any rule against recent events and watch the fired-count before enabling it.
Real-time alerts, triaged.
Every rule match becomes an alert with severity, the triggering detection, and a one-click jump to the footage. Acknowledge individually or clear the board.
- Severity-ranked feed — Info, warning and critical — filter to exactly what needs a human right now.
- Footage in one click — Jump straight from an alert to its clip and the exact event that fired it.
- Acknowledge & resolve — Ack individually or “acknowledge all” — keep the board reflecting live status.
Just ask what happened.
An assistant grounded in your real events, alerts and incidents. “What happened last night?” “Were there any unknown people?” — answered with the footage to back it up.
- Grounded in live data — Answers cite real detections, alerts and incidents — not guesses.
- Your choice of model — Local (Ollama / OpenAI-compatible) for full privacy, or OpenAI, Claude & Gemini for the hard cases.
- Describe-scene on demand — Run the vision model on a live frame to caption exactly what a camera sees right now.
Find every camera on the network.
An ONVIF scan discovers cameras automatically, tests the connection with a live preview, and lets you tune detection per camera — confidence, classes, and zones of interest.
- ONVIF network scan — Auto-discover cameras and verify them with a live preview before adding.
- Zones & line-crossing — Draw regions of interest and crossing lines per camera; count and alert only where it matters.
- Groups & floor plans — Organize cameras by zone or site, and drop them onto a building floor plan as live pins.
Faces you trust. Faces you flag.
Enroll staff and known visitors by name and photo. OricoAI recognizes them across every camera, tags them on events, and flags everyone else as unknown.
- Enroll in seconds — Name plus a photo — the person is recognized everywhere from then on.
- Known vs. unknown — Drive rules off identity — “unknown face after hours” becomes a one-line alert.
- Find a person — Search the timeline for where and when a specific identity was seen across all cameras.
Scale from one site to a portfolio.
Designate one install as the hub and connect edge sites with a pairing code. Each location keeps processing locally; the hub gives you a single view — with licensing controlled per device.
- Hub & edge — Edge sites run autonomously and stay private; the hub aggregates status and alerts.
- Per-device licensing — Authorize each device by its Machine ID — seats are issued and revoked centrally.
- Connect over VPN / tunnel — Edges reach the hub over your own DDNS, VPN or tunnel — no exposure to the public cloud.
Hardened, observable, role-based.
Built to be trusted: scoped access, encrypted secrets, and notifications to every channel your team already uses.
- Role-based access — Admin · Operator · Viewer — scoped control of cameras, rules, alerts and users.
- Encrypted at rest — Camera credentials & API keys sealed with AES-256-GCM and masked in every response.
- Notify any channel — Telegram, Discord, Slack, webhooks & MQTT — plus config backup / restore and an audit log.
On-device intelligence, on your hardware.
Every camera is analyzed locally and normalized into one searchable event stream — no frame ever leaves the building.
| Your hardware | Runtime | Acceleration |
|---|---|---|
| Apple Silicon | CoreML | Neural Engine / GPU |
| NVIDIA GPU | TensorRT | CUDA |
| Intel CPU / iGPU | ONNX | OpenVINO |
| Any CPU | ONNX | native fallback |
The on-prem alternative to cloud camera AI.
Cloud camera services
- Footage leaves your network
- Per-camera monthly AI fees
- Cloud round-trip latency & outages
- Locked to proprietary hardware
- Closed, fixed feature set
OricoAI
- 100% on-prem — nothing leaves the building
- One platform license — no per-camera cloud fees
- Edge-fast response, fully offline-capable
- Any RTSP / ONVIF camera, any GPU / NPU / CPU
- Your choice of AI model, white-label ready
Private AI for every camera you run.
Deploy OricoAI on your own hardware and put an AI security operator behind every camera — in your building, on your terms.