Own Your Security. Bring Your Own AI.

OpenNVR is an offline-first, open-source Network Video Recorder designed for organizations that cannot or will not place their cameras or AI processing in a vendor's cloud. Built on a peer-reviewed security architecture, OpenNVR keeps every video frame and every AI inference on hardware you own, isolates cameras on non-routable networks, and allows you to integrate any AI model as a first-class, governed detector. You retain full control: no cloud accounts, no API keys, and no per-camera fees.

Benefits

OpenNVR is built for entities that cannot outsource trust, ranging from air-gapped government sites to homelab setups. Its architecture ensures that surveillance data and analytics never leave the building unless explicitly configured to do so.

• Offline-First by Default:Cameras operate on isolated, non-routable networks with no path to the internet. Cloud connectivity is an explicit opt-in, which is audit-logged.
• Network Isolation:A hardened middleware gateway is the only point of contact for edge operators and analytics. Camera vendor flaws are contained within the camera itself.
• Decoupled AI Architecture:The video server and AI engine run in separate, isolated processes. If an experimental AI model fails, the core NVR recording functionality remains unaffected.
• End-to-End Audit Trail:Every inference carries a correlation ID linking the alert to the specific model run. Model weights are fingerprinted and polled for drift, with all data landing in an append-only log suitable for regulatory handover.

Use Cases

OpenNVR introduces a voice-driven interface that allows users to interact with their surveillance system naturally.

1. Input:You speak a query, such as "Is there a person at the front door?" Silero voice-activity detection identifies the turn, and Whisper transcribes it locally.
2. Reasoning:A local LLM hosted via Ollama with tool-calling capabilities decides which AI adapter to invoke, such as BLIP for scene captioning, YOLOv8 for object detection, or InsightFace for recognition.
3. Output:The system answers based on a live frame or recorded footage, and Piper synthesizes the reply into speech. The entire loop runs offline on the local hardware.

The platform uses an AI Adapter Contract that allows any model behind a REST or WebSocket endpoint to become a first-class detector.

• Bring Your Own Model:Users can wrap any model in approximately 30 lines of Python using the Apache-2.0 SDK.
• Flexibility:Adapters can be licensed under any terms, including proprietary or classified licenses. The platform core remains AGPLv3, but the SDK allows for closed-source applications.
• Monitoring:Built-in fleet monitoring tracks latency percentiles, outcomes, saturation, CPU/GPU usage, and model-fingerprint drift per adapter.

OpenNVR ships with ten detection apps out of the box, accessible via a one-click app store. Each app features a live dashboard, zone-drawing configuration, and operator actions.

• Perimeter Security:Intrusion detection, loitering detection, line crossing, and abandoned object detection.
• Analytics:Occupancy counting with live per-zone gauges.
• Forensics:Natural language footage search using indexed object labels and scene captions.
• Vehicle & Access:License-plate recognition with live watchlist management and smart doorbell functionality.
• Integration:Home Assistant relay via MQTT discovery.

Pricing

OpenNVR is free and open source under the AGPLv3 license. There are no per-camera fees, per-seat licenses, or cloud subscriptions. A commercial license is required only for businesses that sell hardware with OpenNVR pre-installed under their own brand with proprietary changes, embed it in proprietary software, or host it as a closed service. Self-hosting remains free forever.

Vibes

OpenNVR distinguishes itself through a published open AI adapter contract, an end-to-end audit chain with drift detection, peer-reviewed architecture, and a voice agent that operates entirely on local hardware. Users can pull pre-built images, point the system at a camera, and immediately begin using detections. The platform encourages building custom solutions through its open contract, allowing users to solve specific problems without vendor lock-in.

Additional Information

OpenNVR is designed to meet strict regulatory requirements, including Section 889 compliance for government and defense sectors.

• Encryption:Streams are served over RTSPS, HLS-over-TLS, and WebRTC through MediaMTX. Plaintext is refused on operator-facing transports.
• Access Control:Granular Role-Based Access Control down to the per-camera level, enforced with Multi-Factor Authentication and automatic JWT token rotation.
• Compliance:Provides covered-vendor detection and audit-ready evidence. Roles are strictly defined: Viewers look, Operators arm, and Admins govern.

OpenNVR runs on a wide variety of hardware, including Raspberry Pi 5, laptops, and servers. Pre-built images are available. It is compatible with any IP camera that speaks ONVIF or RTSP, covering virtually all modern brands like Hikvision, Dahua, Axis, and Reolink. Default models include YOLO, BLIP, InsightFace, and LPR.