MIT Licensed · 100% Open Source · No Hidden Components

Decision-Driven AI
Video Intelligence Platform

End-to-end AI fusion for the video IoT field. Solve data silos, fragmented intelligence, difficult customization, and high costs — with private deployment, full-stack controllability, and cloud-edge-device collaboration.

<1.2s End-to-End Latency
500+ Concurrent Streams
17 Export Formats
80+ Target Types
99.99% Availability

Five Problems, One Solution

Traditional video surveillance suffers from fundamental limitations. VLStream addresses all of them.

Data Silos

Unified video aggregation across all protocols — GB28181, RTSP, ONVIF, GA/T1400. One platform for all sources.

Fragmented AI

Dual AI closed loop — discriminative AI for perception plus decision-making AI for execution. See, judge, act.

Hard to Customize

Import TensorFlow/PyTorch models, train custom algorithms, and deploy industry-specific solutions — zero vendor lock-in.

Security Risks

Private deployment, data never leaves your network. GB35114 compliant, encryption at rest and in transit, full audit logging.

High Costs

MIT licensed, no license fees, no per-node charges. Reuse existing cameras with edge AI modules — no hardware replacement.

Dual AI Closed Loop

From perceiving the world to making decisions and taking action — fully automated intelligence pipeline.

Discriminative AI Engine

Perception

See and understand the visual world with industrial-grade accuracy.

  • 80+ target types (people, vehicles, objects)
  • Behavior analysis (loitering, running, gathering)
  • 320,000+ face database with real-time matching
  • 99% license plate recognition accuracy
  • 10+ anomaly types (smoke, intrusion, boundary crossing)
  • Feature extraction (faces, plates, clothing attributes)

Decision-Making AI Engine

Execution

Judge situations and orchestrate automated responses across systems.

  • Event level determination & priority sorting
  • Trend prediction & resource scheduling
  • Multi-channel alarm (PSTN, SMS, APP, Email)
  • Linkage control (PTZ, lights, access gates)
  • Automatic ticket generation & dispatch
  • 1000+ concurrent stream orchestration

Four-Layer Full-Link Architecture

Cloud-edge-device collaboration with dynamic scheduling. Global decisions in the cloud, real-time analysis at the edge, data collection on device.

Platform Layer — VLStream Cloud

Core Hub
Workbench & Dashboard
Video Aggregation Engine
Integrated Train & Deploy
Computing Power Scheduler
Decision AI Engine
API Gateway & SDKs
Multi-Tenant Isolation
IoT Object Model
VLS Protocol / WebRTC P2P / MQTT / REST

Edge Layer — VLStream Edge

Real-Time Intelligence
Edge AI Box (2T/6T/20T)
Real-Time AI Analysis
Local Alarm & Cache
Resume After Network Outage
4G / Ethernet / Wi-Fi

Device Layer — Sensing & Acquisition

End Devices
AI Native Cameras
NVR with Inference
IoT Sensors (Temp/Smoke/Radar)
3rd Party (Hikvision/Dahua)
GB28181 / RTSP / ONVIF / GA/T1400

Application Layer — Client Software

User Interface
VLStream CMS (Desktop)
Mobile App (iOS/Android/HarmonyOS)
Web Console
APIs & SDKs

Purpose-Built AI Hardware

From native AI cameras to edge computing boxes — deploy AI analytics on any infrastructure.

AI IPCamera

Purpose-built intelligent cameras with on-device AI — from entry-level surveillance to flagship multi-model analytics.

Basic

OORT-IPC-4G-30M

Entry Level
  • 4G/Wired dual network
  • AIISP full-color night vision
  • Automatic zoom
  • IP66 weatherproof
  • -30℃ to +60℃ wide temp
Pro

OORT-6600-2.5

1 TOPS AI
  • 2.5T edge computing power
  • On-device object detection
  • Optical fog penetration
  • Smart target tracking
  • Local inference + cloud sync
Flagship

OORT-6600-20

20 TOPS AI
  • 20T edge computing power
  • Full video structuring
  • Multi-stream parallel analysis
  • Complex scene reasoning
  • VLM-ready architecture

AI-Box VLStream Edge

Edge AI computing terminals that retrofit your existing IP cameras — add intelligent analytics via RTSP without replacing your camera infrastructure.

Lite

OORT-BOX-AI-1

1 Channels
  • 1-channel RTSP input
  • On-device object detection
  • Real-time alert push
  • Compact fanless design
  • VLStream Open-source software included
Pro

OORT-BOX-AI-32

32 Channels
  • 32-channel RTSP input
  • Full video structuring
  • AI visual behavior analysis
  • Complex scene reasoning
  • VLM-ready architecture

AI-vLLM-SERVER vLLM Inference

Purpose-built vLLM inference servers — no model training required, direct vLLM recognition out of the box.

Entry

OORT-AI-vLLM-SVR-1

1 Channel / 20 TOPS
  • 1-channel video input
  • 20T computing power
  • No model training required
  • Direct vLLM inference
  • Out-of-the-box deployment
Flagship

OORT-AI-vLLM-SVR-64

64 Channels / 1P
  • 64-channel video input
  • 1P computing power
  • No model training required
  • Direct vLLM inference
  • Out-of-the-box deployment
  • Optional integrated cabinet delivery Built-in liquid cooling technology Edge deployment without dedicated server room

Supported Chipsets

HiSilicon HI3516 Rockchip RK3588 Rockchip RK1106 Ingenic T31/T41 Goke GK720x ESP32 (RTOS) BK7258 (RTOS)

Real-Time VLM Streaming

Let large vision-language models watch video streams in real time — understand scenes frame by frame with streaming output.

1
Video Stream Input

RTSP / WebCam / File at 15-30 FPS. Continuous frame capture with zero drops.

2
Frame Scheduler

Intelligent frame skipping and uniform sampling. Controls flow to prevent AI overload.

3
VLM Inference

LLaVA, Qwen-VL, InternVL — any open-source vision model. No training needed.

4
Streaming Output

Real-time answers in milliseconds. No waiting for video to end.

Supported Models
LLaVA Qwen-VL InternVL Qwen-VL-7B
Use Cases
  • Real-time surveillance analysis
  • Live streaming commentary
  • Robot vision understanding
  • AR/Smart glasses perception
  • Document & invoice OCR
  • Multi-channel parallel analysis
Performance
~200ms Inference Latency
24GB Min GPU (Multi-Channel)

From Raw Data to Production Models

Integrated training and deployment — verify inference in your browser, then one-click deploy to edge devices remotely.

01

Annotate

Built-in labeling for detection, segmentation, and classification. Collaborative workflows with quality assurance.

02

Train

AutoML or full manual control. Distributed GPU training with dynamic resource scheduling. Import TensorFlow/PyTorch models.

03

Optimize

Export to 17 formats: ONNX, TensorRT, CoreML, TFLite, OpenVINO, and more. Quantization and pruning for edge.

04

Deploy

Verify in browser, then one-click remote deploy. OTA updates, A/B testing, rollback, and real-time performance monitoring.

Enterprise-Grade Performance

Built for demanding production environments with industry-leading specifications.

<1.2s
End-to-End Latency
Self-developed VLS protocol, TCP-based, superior to RTMP/RTSP
30%
Packet Loss Resistance
No stuttering at 30% loss, recoverable at 50% packet loss
<200ms
AI Analysis Latency
Real-time inference on edge with hardware acceleration
70%
Bandwidth Savings
WebRTC P2P eliminates relay servers, direct cam-to-client
500+
Streams per Server
Single server handles 500+ video streams, 100+ AI analysis
4K
Max Resolution
H.265 hardware encode/decode, adaptive 128kbps to 8Mbps

Self-Developed VLS Protocol

Purpose-built for video IoT. Low latency, extreme packet loss resistance, and universal compatibility.

Video Transmission

  • VLS Protocol: TCP-based, <1.2s latency, no stutter at 30% loss
  • WebRTC P2P: Millisecond interconnection, 70% bandwidth savings
  • Multi-protocol: HLS, FLV, RTSP, GB28181 — all terminals
  • Encoding: H.264/H.265, 4K/1080p/720p adaptive

IoT & Control

  • MQTT: Transparent command transmission to devices
  • IoT Center: TCP/IP 4G network capabilities for peripherals
  • Object Model: Standardized device descriptions & sync
  • Fusion: Video stream + sensor data combined alerts

Security & Compliance

  • GB28181: National surveillance standard
  • GB35114: Enhanced security certification
  • Encryption: In-transit and at-rest data protection
  • Audit: Full operation logging & access control

Deploy Anywhere

Private cloud, hybrid, or air-gapped. Data never leaves your network. x86 and ARM supported.

On-Premise

Full air-gap support for sensitive environments. Complete data sovereignty on your own infrastructure.

curl -fsSL https://get.vlstream.dev | bash

Docker

One-click deployment with Docker Compose. Full stack running in 5 minutes with GPU passthrough.

docker compose up -d

Kubernetes

Production Helm charts with auto-scaling, HA clustering, load balancing, and automatic failover.

helm install vlstream oort/vlstream

Platform Compatibility

x86 Servers & PCs
ARM Edge & Mobile
GPU NVIDIA / AMD
IPC Industrial Control

Built for Real-World Scale

Deployed across industries where intelligent video analysis drives operational decisions.

Smart City

Integrated video surveillance, urban violation identification, closed-loop work order processing, and emergency command dispatch.

Public Safety

Crowd density monitoring, abnormal behavior early warning, facial recognition, and emergency resource dispatch for large events.

Smart Campus

Suspicious loitering detection, illegal parking, helmet compliance, smoke/fire alerts, and perimeter intrusion monitoring.

Digital Government

Cross-departmental video sharing, unified scheduling, and intelligent analysis for public security, emergency, and transportation.

Industrial & Mining

Equipment fault detection, hazardous area intrusion, PPE violation detection, environmental monitoring, and remote inspection.

Smart Agriculture

Forest fire prevention, river flood control, power line inspection, growth monitoring, and pest identification in weak-network areas.

100% Open Source

MIT licensed. No hidden proprietary components, no closed-source modules. Core source code, VLS protocol, AI engine, and scheduling engine — all open.

VLStream-Cloud MIT

Cloud platform core — management, AI engines, scheduling, video aggregation, and decision-making.

VLStream-NVR MIT

Network video recorder — storage, playback, management, and local inference capabilities.

VLStream-CMS MIT

Cross-platform desktop client (Windows/macOS/Linux) — multi-device monitoring, electronic maps, alarm pop-ups.

VLStream-APP MIT

Mobile app (iOS/Android/HarmonyOS) — live preview, playback, alerts, two-way voice, PTZ control.

VLStream-IPCamera MIT

AI native camera firmware + SDK — supports HiSilicon, Rockchip, Ingenic, and Goke chipsets.

VLStream-Edge MIT

Edge AI box firmware + algorithm packages — real-time analysis, local alerts, computing power caching.

VLS-Protocol MIT

Self-developed VLS protocol — low latency, packet loss resistance, MQTT IoT control, and transparent device commands.

VLStream-Web MIT

Front-end/back-end separated architecture — modern codebase, clean API interfaces, easy to customize and build your own product on top of VLStream.

Why VLStream vs. Closed Source

Compare VLStream with traditional closed-source video platforms.

Dimension VLStream (Open Source) Traditional Platforms
Source Code 100% open, no hidden components Closed source, core not visible
Commercial Cost Free commercial use, no license fees High licensing, annual, and customization fees
Customization Deep secondary development, full refactoring Long cycle, high cost, vendor lock-in
Data Security Private deploy, data never leaves network Cloud upload, data leakage risk
Protocol National standards + self-developed, 100% reuse Proprietary protocols, hard to reuse
AI Control Full AI engine control, custom algorithms Closed algorithms, limited upgrades

Three Core Advantages

01

Reuse Existing Equipment

Existing cameras and IPCs stay in place. OortCloud edge AI modules give them intelligence — no hardware replacement needed. Maximize ROI on existing infrastructure.

02

Human to AI Monitoring

Transform manual surveillance into AI-powered monitoring. Built-in alert push sends real-time notifications to relevant personnel. Reduce labor costs and create perceptive intelligent agents.

03

Configurable AI Algorithms

Train and dynamically load algorithms on demand. Different scenarios, different time periods, different early warning items — all configurable without code changes.

Ready to ship Vision AI?

Deploy VLStream in 5 minutes. Transform your video infrastructure with decision-driven AI. 100% open source, zero vendor lock-in.

Need commercial support? OortCloud offers paid technical support, custom development, and deployment services.