Self-host OpenNVR

OpenNVR is an open-source network video recorder that handles ONVIF and RTSP camera streams, records continuously, and serves live video via WebRTC and HLS. It includes an AI platform where any model — object detection, license plate reading, face recognition, pose estimation, or vision-language models — connects through a published open adapter contract. Twelve built-in apps provide intrusion alerts, loitering detection, occupancy counting, tripwire counting, abandoned object alerts, license plate recognition with gate control, a smart doorbell with face enrollment, package delivery tracking, guard scan compliance, plain-language footage search, and alert notifications via Telegram, ntfy, Discord, Slack, email, and webhooks. A voice agent lets you speak to cameras and hear answers from a local LLM. The design enforces offline-first operation: cameras sit on a segregated network without internet access, two default-deny gates block cloud egress, every inference carries a correlation ID and model fingerprint into an append-only audit log, and a covered-vendor self-check is included. It runs on hardware from a Raspberry Pi 5 up to GPU servers, installs via Docker Compose with a guided launcher, and integrates with Home Assistant through MQTT discovery. The core uses AGPL-3.0-or-later; the app and adapter SDKs are Apache-2.0 so extensions can be proprietary. A commercial license covers OEM embedding and hosted offerings.