OpenNVR
100% Open-Source

Own Your Security.
Bring Your Own AI.

The world's first mathematically air-gapped NVR built for infinite AI scaling. Securely isolate your cameras on a zero-trust network while plugging into over 100,000+ Hugging Face models instantly.

Get Started Read Documentation
infer_huggingface.sh
# Send secure frame to Hugging Face adapter
curl -X POST http://opennvr.local/infer \
-H "Content-Type: application/json" \
-d '{
"task": "image-to-text",
"input_data": {
"model_name": "Salesforce/blip-image-captioning-base",
"inputs": { "image": "opennvr://camera_0/latest.jpg" }
}
'
Response:
{ "label": "A person walking near the secure gate." }

Enterprise Hardened,
Open Source Core.

We didn't just build an NVR. We built a zero-trust network appliance that happens to stream high-definition video perfectly.

Zero-Trust Architecture

Cameras are placed on non-routable VLANs. The NVR mathematically guarantees no unauthorized access to your hardware feeds.

Offline-First Edge Compute

Run advanced YOLOv11 and InsightFace biometric tracking entirely on your local hardware. No cloud pinging, no latency, no subscription.

Integrated Suricata IDS

Built-in Intrusion Detection System actively scans internal network traffic for malware lateral movement and unauthorized scanning.

Encrypted Multi-Relay Streaming

View hundreds of streams securely over poor connections using automatic multi-relay WebRTC scaling and hardware-accelerated MediaMTX.

Granular RBAC + Auth

Enforce strict Role-Based Access Control down to the per-camera level. Includes mandatory MFA, audit logging, and automated JWT rotation.

Decoupled AI Architecture

The video server and the AI engine run in separate, isolated processes. If an experimental AI model crashes, your cameras never drop.

100,000+ Models.
Zero Code Required.

Open-NVR ships with its own isolated AI Adapters Engine. Want edge performance? Run our optimized YOLOv11 and InsightFace handlers locally in full offline mode.

Need cloud-scale reasoning? Tap directly into the Hugging Face Inference Network. Switch from local bounding boxes to full Vision-Language Scene Description by changing a single parameter in your NVR dashboard.

  • Bring Your Own Model (BYOM) via ONNX or PyTorch
  • Native Hugging Face API Integration
  • Fault-Tolerant, Crash-Proof Plugin System
Explore Hugging Face
adapter/tasks/my_custom_ai/
HOT RELOAD
__init__.py
schema.json
task.py
class Task(BaseTask):
name = "fire_detector"
def run(self, image, params):
return self.model.predict(image)
✓ Successfully Loaded Plugin: fire_detector