Fugleramme's Show HN post reached 573 points in its first 59 minutes on September 15. The project behind that rush is deliberately slow: a Raspberry Pi listens for garden birds, classifies their calls locally, and changes a six-color e-ink collage only when the set of species changes. The Hacker News thread drew fast attention. The repository explains the mechanism. Its software accepts the pace of paper instead of making the screen behave like a tablet.
The frame runs from its maker's kitchen window in Bergen, Norway, with a public web mirror showing the birds heard there. Each supported detected species appears as a cutout from a historic natural-history plate. An empty listening window leaves a bare perch. The project currently contains more than 800 cutouts covering over 400 species, and its README says the artwork is human-made, although AI was used to retouch some scans. That combination gives the machine a visible job without turning every detection into a notification.
A five-second poll meets a 35-second screen
BirdNET-Go handles the microphone input and classification. Fugleramme queries its REST API, maps each scientific name to an illustration, packs the selected birds onto a page, and sends the result to both an e-ink panel and a browser kiosk. The service loop polls every five seconds, yet it prepares a new panel image only when the render key changes or a configured refresh becomes due. If a panel push fails, the pending image survives for another attempt instead of being rendered again.
That restraint matches the display. Pimoroni lists the 13.3-inch Inky Impression as a 1,600-by-1,200-pixel Spectra 6 panel with red, green, blue, yellow, black, and white particles. Its quoted core refresh is 12 seconds, while a complete real-world cycle can take 20 to 35 seconds depending on the Pi and panel. Electronic paper also draws power during refresh rather than continuously, according to the manufacturer's specifications. A five-second detector loop therefore cannot become a five-second display loop, and Fugleramme does not try to force it.
The composition carries more data than a row of thumbnails would. Fugleramme scales birds using body-mass data, gives larger species more weight near the center, and can place translated labels beneath the cutouts. The collage renderer also keeps artwork choices stable within a display window. That deterministic layout reduces pointless visual churn and makes a mixed flock readable at a glance.
A panel is optional. The project README documents a web-only container that can connect to BirdNET-Go elsewhere on the network, while the kiosk serves the same composition in full color. That path lets a developer test detection, species matching, and layout on an existing computer before buying the A4-sized screen.
The model sees farther than the art library
BirdNET-Go's default embedded BirdNET v2.4 model lists more than 6,500 bird species, and the upstream project can add other classifiers for wildlife and bats. Fugleramme can currently draw a little over 400 species. In many locations, illustration coverage will run out long before the classifier's label set does. The contrast between the detector's reach and Fugleramme's art catalogue is the main maintenance job for anyone building one outside its strongest regions.
Those strongest regions are the Nordics, the British Isles, and Germany, according to the repository. Elsewhere, the admin page marks a detection as having no art, and the service logs the missing species. The artwork contribution guide asks contributors to find a public-domain or compatible open image, cut the bird out manually, preserve attribution, preview it against the frame's paper, and then let a project tool encode it as WebP. Geography expands one carefully sourced plate at a time.
The stack has several licenses. Fugleramme's code is MIT-licensed, while its classic artwork set uses CC BY-SA 4.0. BirdNET-Go and its detector are installed separately under CC BY-NC-SA 4.0, which restricts commercial use. The project also credits CC BY 4.0 AVONET data for bird sizes in its license notes. Anyone planning to sell a derivative needs to review each layer instead of treating the repository's MIT badge as the license for the whole stack.
The LAN is part of the design
BirdNET-Go describes its default operation as local-only and makes optional Sentry telemetry opt-in. Fugleramme can read detections from a BirdNET-Go instance on the same Pi or another machine on the local network. Daily classification can therefore stay inside the home, while the operations guide says release checks and software updates require an internet connection. The separation is useful for a microphone that may run continuously near a house.
The current web service assumes a trusted network. A comment in Fugleramme's server implementation states that both the kiosk and admin views have no authentication because they are intended to be open on the LAN. The installer defaults the frame to port 8080 and its bundled BirdNET-Go instance to 8090. Builders should keep those services on a trusted network and avoid public port forwarding unless they add a properly configured access layer. This is a deployment constraint in the current design. The repository does not report it as a vulnerability.
The full hardware recipe is specific. The hardware guide recommends a 2 GB Raspberry Pi 5, at least 32 GB of storage, an active cooler, the 13.3-inch panel, a microphone, a USB audio adapter, and the official power supply. The author calls the cooler mandatory when BirdNET-Go is classifying continuously and says microphone choice and placement matter more to detection quality than buying a larger Pi. A frame-only client connected to an existing detector can use a Pi 4 or Pi Zero 2 W.
Setup still asks for comfort with a Linux box. The installation guide uses a shell installer that clones the repository, installs dependencies, enables SPI and I2C, creates a systemd service, and usually finishes with a reboot. It prompts before making changes. A -y flag accepts all defaults. As with any installer piped from the network, a cautious builder should download and inspect the script first. The project labels itself early-stage, so that inspection is also a chance to understand what future updates may alter.
What 573 points are measuring
The Hacker News questions focused on the build itself. One reader asked how much work Texas artwork would require, and another wanted a browsable path back to the original plates. In the thread, the maker replied that a local collection may need only a few dozen frequent visitors and pointed people to the repository's attribution manifests. Those replies measure interest and show where prospective builders immediately found unfinished work. Coverage itself must be checked against the repository.
Fugleramme gives each layer a bounded role: BirdNET-Go identifies a sound, Fugleramme chooses and arranges the art, and e-ink holds the result until the garden changes. The result is easy to inspect because the output is physical and the failure states are visible. A missing illustration leaves a named gap. A slow screen dictates a slower update policy. In an unsupported region, the next task is concrete: curate the missing plate. The artwork workflow documents that job.
The release history shows what progress looks like here. Version 0.21.3, published on September 14, added six species, removed a duplicate plate, evened the paper beneath 233 images, and reworked 23 other scans, according to the release notes. The next useful signal is geographic: which gardens can the frame actually draw? Build reports from outside Europe will show how portable the project really is.