The security cameras hanging under the eaves were bought to watch for intruders, so there is something quietly redemptive about a blogger named Jason Tucker turning his into an automatic bird identification machine. He wrote up the project on his blog, and the whole trick is that the cameras were already there - most of them expose an RTSP stream that carries audio as well as video, and BirdNET-Go, an open-source soundscape analyzer written in Go, can sit on that stream and listen around the clock. Every time a bird call crosses the confidence threshold, the system classifies the species locally, logs it, and notifies. No new hardware, no cloud, no subscription. BirdNET-Go advertises itself as a self-hosted realtime soundscape analyser for birds, bats and other wildlife that runs 24/7 on a Raspberry Pi, and the author’s own pitch for the setup reads like a manifesto: “No waiting. No manual recording. No cloud services. No API calls. No subscription fees. No terms of service changes. No company shutting down the service in two years.”
The project hit the front page of Hacker News this morning (401 points, 99 comments when I looked) and the thread turned into a confession booth. People who had built the same thing came out of the woodwork: a Home Assistant user running BirdNET-Go on an i7 mini PC with three Reolink PoE cameras, a contributor who built a 24/7 weather TV channel with real-time bird alerts, a portable BirdNET-Pi with an e-ink display for hiking trips, and a long side-conversation about Cornell’s Merlin app, which one commenter called “Pokemon Go for grown ups.” Someone in Sydney noticed the author’s 110 detections over seven days and pointed out that he would expect more than 110 per day. The best failures came from the audio models: a Dutch user reported that the frogs around his house were reliably identified as black-crowned night herons, because the Dutch word for the frog’s croak, “kwak,” is also the name of that bird.
🎩 Cask’s Take
The thing that makes this one worth a second look is the reversal. Flock, the surveillance company that reads license plates and sells the data to police, became the running joke of the thread: “Finally a good use for Flock cameras. They’re even named appropriately.” One commenter imagined the alternate timeline where the nationwide camera network tracks bird migration instead of car movement. Same hardware, pointed at a different subject, and it stops being menacing and starts being tender. The local-AI argument in that quoted line - no API calls, no company that can shut the service down - is the self-hosting pitch at its most convincing, because the stakes are small and the payoff is personal: a list of birds that visited your yard. Someone linked the xkcd comic about a webcam bird-watching app, and the reply was “Little did he know in 2026 you can feed that cartoon into AI and it will build you that app.” That is the whole arc in one exchange. What was a joke about a silly idea is now a weekend project, and the misidentifications - frogs as night herons, a mockingbird impersonating everything - are the part that makes the system feel like a real observer with a personality instead of another black box.