Unlike cloud NVRs (Ring, Nest, Verkada), OpenNVR is 100% Open Source and runs entirely on hardware you own — footage never leaves your network. Unlike fixed-stack NVRs, the AI is pluggable: any model becomes a detector via an open contract (~30 lines of Python) — YOLOv8, license plates, faces, or your own. And you can literally talk to your cameras: ask a question, a local LLM checks a live frame and answers out loud. No cloud, no subscriptions, no per-camera fees. Peer-reviewed architecture.
No reviews yetBe the first to leave a review for OpenNVR
Maker
📌
Hi Product Hunt! 👋 Builder here.
What inspired this: IP cameras have a trust problem. In 2016, hijacked cameras formed the botnet that broke the internet; in 2021, one breach exposed ~150,000 live feeds from hospitals, schools, and prisons. And the industry's answer to "smart cameras" has mostly been: stream everything to our cloud, pay a subscription, trust us.
Coming from years of building real-time voice and video systems, we wanted the opposite — so we built OpenNVR: 100% Open Source, offline-first AI video surveillance that runs entirely on hardware you own.
🔒 Cameras live on an isolated network they can't escape. Default-deny gates block all cloud/AI egress until you open them.
🧩 Any AI model plugs in — YOLOv8, license plates, face recognition, tracking, speech — through one open contract. Wrap your own model in ~30 lines of Python.
🎙️ And the part I love most: you can talk to your cameras. Ask "is there a person at the front door?" — Whisper transcribes, a local LLM checks a live frame with YOLOv8, and Piper speaks the answer back. Zero cloud involved.
It's AGPLv3 core + Apache-2.0 SDK, ships with 13 example apps (10 one-click), runs via Docker Compose in ~5 minutes, and the architecture is published as a peer-citable paper — auditable infrastructure, not a black box.
Would love your feedback — I'll be here all day answering questions. What would you point it at first? 🎥
Report
💎 Pixel perfection
Love that the detector contract is literally a Python file you can drop in — makes swapping in a custom YOLO model feel like swapping a lens instead of wrestling with vendor SDKs. The fact that everything still happens locally is just icing.
Report
Maker
@sabrihacar Thanks for reviewing this and yes it is a design which gives AI models and developers endless possibilites.
Report
the pluggable detector contract is genuinely clever, thirty lines of python to drop in any model feels like the right level of openness without drowning people in complexity
Report
Maker
@ebubekird2os Thanks for reviewing this and yes it is a design which gives AI models endless possibilites.
Love that the detector contract is literally a Python file you can drop in — makes swapping in a custom YOLO model feel like swapping a lens instead of wrestling with vendor SDKs. The fact that everything still happens locally is just icing.
@sabrihacar Thanks for reviewing this and yes it is a design which gives AI models and developers endless possibilites.
the pluggable detector contract is genuinely clever, thirty lines of python to drop in any model feels like the right level of openness without drowning people in complexity
@ebubekird2os Thanks for reviewing this and yes it is a design which gives AI models endless possibilites.