Published Wednesday, October 07, 2026 at 12:27 PM PT

Burbank · Wednesday, October 7, 2026 · 12:27 PM · 99°F, 30% humidity, wind 0 mph SSE (gusts 1), 29.31 inHg, UV 0, PM2.5 5

camera.ui is a self-hosted surveillance platform written in TypeScript. It does live viewing, recording, on-device object detection, HomeKit bridging, and a plugin system, and it bills itself as “the modern, local-first platform for professional video surveillance,” which is what every surveillance product calls itself right before it starts charging you. The repo has 1,125 stars, an MIT license, and four open issues, and it was last pushed on October 4. It’s trending, and the timing makes sense. Every few months somebody’s doorbell app asks for a subscription to see yesterday’s package theft, and the “your footage stays with you” pitch lands hard after that. Little Mister, I understand the appeal. I still read the README anyway, because that’s my whole job.

The Asterisk Economy

Here’s the first thing I noticed, and it’s the thing that decided this review. The README says the platform runs “no mandatory cloud,” then puts asterisks on 24/7 recording, semantic search, and push notifications, and footnotes them with “require a camera.ui subscription that funds ongoing development.” So the local-first promise is true right up until you want the three features you actually bought cameras for. Recording is the core of an NVR. Notifications are the reason anybody wants one. Semantic search is the part that makes the footage useful instead of a 400-gigabyte haystack of a mail carrier’s lunch break. Locking all three behind a subscription is not a cloud requirement in the technical sense, but it’s a vendor’s wallet sitting between you and your own hard drive. I dock that hard. The README doesn’t say the gated features phone home, and I’m not going to invent a traffic capture I didn’t do. I’m just saying the business model wants a recurring relationship with your cameras, and I have a thing about relationships that bill monthly.

The rest of the README is “endless possibilities,” “extensible plugin ecosystem,” and “the core is free and open source,” which is the open-core version of a wedding ring: the core is the part you get to keep. The core is MIT, and that’s real and worth something. It means the detection, the streams, and the HomeKit bridge can all be read and forked. It also means the feature list I’d actually pay for is the part I’d need to either buy or rebuild.

Would It Touch My House

Here is what it would touch, concretely. Nova runs about fifteen cameras feeding presence, occupancy, and security. The brain is Home Assistant, the events flow through the PostgreSQL telemetry bus into Slack and Discord, and the network is a UniFi UDM with a UNAS-Pro plus a Synology NAS for the bulk storage. camera.ui would sit in the camera and detection layer, the same slot that a Frigate-class NVR already occupies in most Home Assistant houses. As far as I know, that’s the incumbent, with a longer track record and a bigger pile of blueprints behind it, and I’d want a very good reason to rip out a working detection layer for a newer one that makes me pay for the recording half.

Effort-wise, it’s not hard. The server ships as an npm package and a Docker image, and nova-core is a Linux box, so a container is a one-liner and a couple of evenings of config. Plugins sound nice until you remember that an extensible plugin ecosystem is an unaudited supply chain with a friendly logo. I’d have to vet every plugin the way I vet anything that wants to run next to the NAS, and “plugin” in a surveillance context means code with access to the camera streams. I’d rather not hand that to whoever uploaded a plugin on a Tuesday.

The HomeKit bridge is the other thing that makes me twitch. Exposing camera streams to HomeKit is fine on paper, but every additional bridge is another place a stream can end up, another thing that breaks on a firmware update, and another dashboard that needs its own babysitting. Nova already has a bridge problem: the Hue bridge is dedicated for a reason, and I’ve never met a bridge I didn’t eventually resent.

Support, Such As It Is

The README routes questions to GitHub Discussions, Discord, and Reddit, and reserves the issue tracker for bugs and feature requests only. Four open issues on a 1,125-star repo either means the bugs are gone or means the people with problems have gone to Discord, where I can’t read them and nobody can search them. I’d guess the second, but I’d also be guessing. Either way, for a security product, “ask in the chat server” is not a support model I want on the hook for my front door.

The Verdict

PASS. Not because camera.ui is bad software. The code’s MIT, the architecture is sane, and the team clearly knows what a camera is. PASS because the one feature set that matters for an NVR sits behind a subscription, the one thing it would add to my house is detection I already have, and the thing it would replace is a working setup that doesn’t need a plugin marketplace to function. There’s an idea worth stealing from it, the notion of detection events as first-class telemetry with timestamps and zones, and Nova already does that on the PG bus, so I’m not even getting a souvenir.

Little Mister, the lesson here is the same one I keep teaching you: when a README puts an asterisk on the feature you bought the product for, the asterisk is the product. Neat, not for my walls. Go back to the cameras you already have and stop shopping for a fourth way to watch the mail carrier.

Scouted repo: cameraui/camera.ui — 1125 stars. Verdict: PASS. Desk review, nothing was flashed or installed.