Sentinel
AI-Powered Video Management and Command-and-Control Platform
GPU-accelerated video management with live surveillance, face recognition, person tracking, ANPR, investigations and evidence management, deployable on a single Ubuntu GPU server, on-premises or in the cloud.
- Live multi-camera surveillance with real-time AI overlays
- Face recognition, cross-camera person tracking and number-plate recognition with watchlists
- Cases, evidence vault, forensic timeline and automated reports
- Runs on your own GPU server — on-premises or in your cloud account
Overview
Sentinel brings video management, AI video analytics and investigation tools into one platform. Cameras stream in over RTSP or are discovered through ONVIF; GPU-accelerated microservices detect objects and faces, track people across cameras, read number plates and flag behaviour; and operators work from a single web console with live walls, a situation room, event feeds, cases and an evidence vault.
Sentinel v8.0 is packaged as Docker services with a one-command installer and runs on a single Ubuntu server with an NVIDIA GPU — bare metal on-premises or a GPU instance in AWS or Azure. A local language model provides the VedaAI assistant without sending data to external AI services.
Key capabilities
-
Live surveillance
Multi-camera HLS streaming with real-time AI overlays and a four-camera situation-room command display with live event feed.
-
Object and behaviour analytics
GPU-accelerated YOLO object detection with behaviour analysis and anomaly detection on live video.
-
Face recognition
Face detection, enrolment, search and matching against a face watchlist (InsightFace models, 512-dimension embeddings).
-
Person tracking
Cross-camera person re-identification.
-
Vehicle analytics and ANPR
Automatic number plate recognition with plate watchlist matching.
-
Investigations and cases
Case management with notes, linked events and an investigation workspace.
-
Evidence and forensic timeline
Evidence vault, recorded-video playback and a forensic timeline of events.
-
Floor plans and zones
Zone maps for sites and floors.
-
Reports
Automated report generation, PDF-ready.
-
VedaAI assistant
Built-in assistant powered by a local LLM through Ollama.
-
Alerts and automation
Event streaming between services over Kafka, with alerting by email and webhooks.
-
VMS and ONVIF integration
ONVIF camera discovery and VMS integration through a dedicated connector service.
-
System health
Live CPU, RAM, disk and GPU metrics, diagnostics and Grafana dashboards.
-
Hot, warm and cold storage Source-reported
Tiered storage lifecycle for recordings and evidence.
Architecture
Sentinel runs as Docker services behind an nginx front end. A Go API gateway handles authentication, REST and WebSocket APIs; Python AI microservices perform inference on GPU; MediaMTX ingests RTSP and serves HLS to browsers; PostgreSQL, Redis, Kafka, Milvus and MinIO provide data, events and storage.
-
Operator console
- Single-page web application
- Live wall and situation room
- Cases, evidence, reports, settings
-
Edge and API
- nginx with TLS 1.2 / 1.3
- Go API gateway: 50+ REST endpoints
- WebSocket live event feed
-
Video
- MediaMTX RTSP ingest and HLS output
- Recording to disk
- ONVIF / VMS connector
-
AI services
- AI inference (YOLO)
- Face recognition and face search
- Person tracker (re-ID)
- Event intelligence
- Report engine
-
Data and messaging
- PostgreSQL
- Redis
- Kafka
- Milvus vector database
- MinIO object storage
-
Intelligence and observability
- Ollama local LLM (VedaAI)
- Prometheus / Grafana
- Audit logs
Technical specifications
| Operating system | Ubuntu 22.04 LTS or 24.04 LTS |
|---|---|
| Packaging | Docker 24+ with Docker Compose v2; one-command installer |
| Minimum | 8 CPU cores, 16 GB RAM, 100 GB SSD, NVIDIA GPU with 8 GB VRAM, 100 Mbps network |
|---|---|
| Recommended | 16+ cores, 32 GB RAM, 500 GB+ NVMe for recordings, NVIDIA T4 / A10G class GPU, 1 Gbps network |
| Drivers | NVIDIA driver 525+ and NVIDIA Container Toolkit 1.14+ |
|---|
| Ingest and delivery | RTSP in (MediaMTX), HLS to browsers |
|---|
| Models | YOLO object detection; InsightFace face recognition; local LLM via Ollama |
|---|
| Stores | PostgreSQL, Redis, Kafka, Milvus, MinIO |
|---|
| Interfaces | REST (JWT bearer authentication) and WebSocket event stream |
|---|
| Tested | Ubuntu bare metal; AWS EC2 g4dn.xlarge (NVIDIA T4) |
|---|---|
| Compatible | AWS EC2 g5.xlarge (A10G); Azure NC6s v3; on-premises NVIDIA servers |
Deployment
-
On-premises
Bare-metal Ubuntu server with an NVIDIA GPU.
-
Cloud (your account)
GPU instances in AWS or Azure.
Requirements
| Server | Ubuntu 22.04 / 24.04 LTS with an NVIDIA GPU (8 GB VRAM minimum) |
|---|---|
| Software | Docker 24+, Docker Compose v2, NVIDIA Container Toolkit, FFmpeg, Python 3.10+ |
| Cameras | RTSP-capable IP cameras; ONVIF for auto-discovery |
| Storage | SSD/NVMe sized for the recording retention you need |
Supported integrations
-
ONVIF cameras
Cameras
Auto-discovery through the VMS connector.
-
RTSP streams
Video
Any RTSP source can be pulled by MediaMTX.
-
Email (SMTP) and webhooks
Alerting
Event notifications to people and systems.
-
Grafana / Prometheus
Observability
Metrics dashboards.
-
REST and WebSocket API
Developers
Integrate events, cameras, cases and watchlists.
Security & governance
-
Authentication
JWT authentication with short-lived tokens and refresh tokens; bcrypt password hashing; forced change of initial passwords.
-
Role-based access
Role-based access control for operators and administrators; MFA-ready.
-
Audit trail
Write operations recorded in an audit log.
-
Transport security
TLS 1.2 / 1.3 on the web front end.
-
Rate limiting and WAF
Per-IP rate limiting and web application firewall rules.
-
Local AI
Analytics and the VedaAI assistant run on the server's own GPU; no external AI service is required.
Use cases
-
Public safety
Command-and-control rooms
Live walls, situation room and event feeds for operators.
-
Law enforcement and security teams
Investigations
Cases, evidence vault, playback and forensic timeline.
-
Traffic and access control
Vehicle monitoring
Number-plate recognition with plate watchlists.
-
Critical infrastructure
Critical sites and campuses
Face watchlists, cross-camera tracking and behaviour alerts.
Documentation & developer resources
Overview
Deployment
Integration
Frequently asked questions
Can Sentinel run on-premises?
Yes. Sentinel is designed to run on a single Ubuntu server with an NVIDIA GPU on-premises, or on GPU instances in your own AWS or Azure account.
How are cameras connected?
Cameras stream over RTSP into MediaMTX, and ONVIF cameras can be discovered automatically through the VMS connector. Browsers receive live video as HLS.
What hardware do we need?
At minimum 8 CPU cores, 16 GB RAM, 100 GB SSD and an NVIDIA GPU with 8 GB VRAM; 16+ cores, 32 GB RAM, NVMe storage and a T4 / A10G-class GPU are recommended. Requirements grow with camera count.
Items marked Source-reported are as stated in VedaCLOUD.AI product material and have not been independently verified.