Hardware guide

The standard set is a single solar 4G box on one post: The kit: Life node v1. It needs no farm WiFi, no LoRaWAN gateway and no monthly subscription, because species recognition runs inside the box and the soil probes are wired into it. This page is the wider market scan behind that choice, plus the alternative sets for the places where the node is not the right answer: no signal at all, or WiFi already at the field edge.

Prices were checked live on 2026-09-12, and the ready-made devices at the bottom of the page on 2026-09-13. They include VAT where the shop shows it. They will drift; the repository is the place to correct them. Everything here was chosen on one criterion: install once, leave it for years, data comes out without a visit. Where that is not possible yet (offline recorders), the guide says so.

The alternative sets

The node covers both streams at once. These three sets each cover one stream, and each answers a situation the node does not. They are named, not lettered, so that a set means the same thing on every page.

A first installation for one farm (three fields, say 20 ha) along the distributed route: 2× connected or offline sound, 3× SE01-LB, one indoor gateway on the farmhouse window. Roughly €1,200 to €1,400 once, under €15 a month connectivity. That is under €10 per hectare per year over five years, before any satellite cost, which is zero. The same farm along the node route is one node per field at about €809, so more money up front for three fields, and in exchange no access point, no gateway, no router, no subscription, and nothing that stops working when the farm changes its WiFi password. The node is the better fit the moment the field is out of WiFi reach, which is most fields; where WiFi does reach, a PUC is the simpler choice (the comparison on the kit page counts both, visits included).

Sound recorders

Offline (SD card)

Cornell's Swift is discontinued; Bugg (Imperial College) is not currently for sale and needs a 50 to 100 W solar kit because it uploads full audio. Neither is recommended today.

Connected (real-time detections, no card swap)

There is no commercial LoRaWAN or NB-IoT recorder that ships species detections today. That remains DIY (Pi + LTE dongle) or 2027 (Wildlife Acoustics cellular module).

Ultrasonic (bats, 192 kHz and up)

Insect cameras (instead of traps)

• DIOPSIS (Faunabit, NL): camera with LED screen, 4G, Naturalis recognition model; ran at 120 sites 2021 to 2024, now in regular monitoring. Price on request: faunabit.eu.
• Insect Detect (open source, DE): OAK-1 camera + Pi Zero 2 W, solar, BOM €445 to €660, GPLv3: docs.

In the oracle design, birds and bats are the routine insect sensor; an insect camera is a good add-on for a subset of places and a fine triggered-visit instrument.

Software that turns audio into data

The licence line matters: if the oracle is ever sold as a service, the BirdNET family needs a commercial licence or a replacement. Perch and scikit-maad do not.

For bats the licence question has an answer today. BattyBirdNET, which BirdNET-Go runs out of the box, is CC BY-NC-SA 4.0 and works on BirdNET's own embeddings, and BatDetect2 is CC BY-NC 4.0, so both are for research and the bench. BSG-BAT is CC BY 4.0: free to use in a paid service with attribution (checked on Zenodo 2026-10-07). Its 21 species cover most of the Dutch list (common, Nathusius' and soprano pipistrelle, noctule, Leisler's, serotine, Daubenton's, both long-eared bats, parti-coloured, barbastelle) but not the other Myotis, and so not the pond bat, for which the Netherlands holds an internationally important share. It wants mono audio at 384 kHz, which is exactly what the AudioMoth USB Microphone gives.

Since release 20260823 BirdNET-Go also runs Google Perch v2 beside BirdNET v2.4 (model files, Apache-2.0, checked 2026-10-05). The ARM build cut to Central Europe is 46 MB, 873 classes (675 species plus 198 sound events such as engines and voices) and about 250 MB of RAM, so it fits a Pi 4 with 2 GB next to BirdNET. Building the image with PERCH_REGION=central-europe enables both, and every detection then names the model or models that heard it (birdnet-go:birdnet+perch). That is the bench comparison. It does not yet clear the licence: the weights are free for commercial use, but BirdNET-Go itself is CC BY-NC-SA 4.0, so a paid service would also need a different runtime around the model.

Soil probes

Recommended: LoRaWAN moisture + temperature (+ EC)

For the oracle's Cycling reading the probes give moisture at two depths, with temperature: from them and open weather data the oracle reads how much water living roots draw from the top 40 cm on dry days (the soil's water loop), so the Sensoterra (no temperature) and the cheap analog probes (Pino-Tech SoilWatch 10 €21 to €28, Ecowitt WH51 €20 to €27 with WiFi gateway, Catnip €12 open hardware) are second choices: fine as extra witnesses, not as the one probe per field.

Soil CO2 (flux) is not in the standard kit

Purpose-built flux instruments (Eosense eosFD, LI-COR LI-8250) are quote-only research equipment, mains or large solar, with moving parts or annual membrane changes. No LoRaWAN buried CO2 probe exists as a product; the one long field trial of buried Seeed CO2 nodes saw 20 units fall to 1 survivor in four years (humidity). The Cycling reading therefore runs on the moisture probes and open weather data (the soil's water loop); a CO2 probe is an optional extra source whose flux the oracle normalises for temperature and moisture (see the specification 3.5). If you want one anyway, the honest budget route is a Sensirion SCD30 in a membrane housing with desiccant (~$60 module) as published in Nguyen and Levintal 2025 (doi 10.5194/soil-11-639-2025).

LoRaWAN gateways and networks

Check ttnmapper.org first: if a community gateway already covers the fields, you need none.

On flat farmland expect 3 to 8 km from a mast-mounted outdoor gateway, 1 to 2 km from an indoor one behind glass.

Network server. The Things Stack Sandbox is free for small, non-commercial use (fair use: 30 s uplink airtime per device per day, plenty for a probe sending every 15 to 30 minutes at SF7 to SF9). The Things Stack Cloud Discovery tier is free for 10 devices with an SLA; Standard is €190/month for 1,000 devices. In the Netherlands KPN LoRa is nationwide: Explorer plan €1.33/month per device, monthly cancellable, devices must be on KPN's qualified list. Helium costs about $0.00001 per uplink but has no roaming with TTN and needs its own network server.

NB-IoT alternative where there is no LoRaWAN: Dragino SE01-NB (€133 at Antratek) with a 1NCE SIM (€13 for ten years, 500 MB).

Ready-made alternatives: can you skip the build?

Checked live on 2026-09-13. Prices include VAT where the shop shows it. The short answer: for either half of the node, nearly. For the whole node, no. Nobody sells it.

Listening, assembled

Not in shops, and worth knowing: Bugg v4 (3G/4G and off-grid solar, inference in the cloud; its GroupGets pre-order has closed) and RFCx Guardian 3 (cellular and solar, in programme deployments only).

The PUC is the strongest of these and this oracle already polls it hourly by station id. Its one constraint is the one that produced the node: WiFi and nothing else, and most fields have none.

Two open reference designs are worth reading before anyone builds a v2 board: Microsoft SPARROW (Jetson Orin Nano with Starlink, solar, edge inference) and Bird@Edge (ESP32 microphone nodes reporting to a Jetson base, about €110 in parts). Neither is a product.

Soil, assembled

Farm21's FS31 is the one product that could replace the soil half of the node outright: cellular built in, no gateway, about a year on a charge, three moisture depths in one probe, and a published REST API. Against two €26.90 wired probes it costs about €280 more up front and €89 a year per node, and it places a company between a place and its own readings. That is the trade, and it is a real one.

The combination is what is missing

No product hears, reads the soil, carries both over one cellular link, and posts the result to a registry the landowner and anyone else can read. The nearest buyable approximation is a PUC plus an FS31, roughly €664 plus €89 a year, and it stops at the first field without WiFi. Adding a 4G router to rescue it rebuilds the €1,110 distributed set that the kit replaced.

This is not a gap in engineering. Every part of it is available and cheap. It is a gap in what the market is for: these devices are sold to deliver data into the seller's platform, so nobody has a reason to build the one that delivers it into someone else's. That is the reason to build it rather than buy it, and it is recorded with its evidence in Prior art.

One thing worth buying anyway: a PUC next to the first node, as an independent BirdNET implementation to check our own detections against. €289 answers the first question any auditor will ask.

Where to buy

• Veldshop.nl (NL): all acoustic recorders above, same-day shipping across the EU.
• LabMaker (Berlin): Open Acoustic Devices' EU distributor (AudioMoth, µMoth, HydroMoth).
• Antratek.nl (NL): Dragino full line, Seeed SenseCAP, gateways.
• iot-shop.de (DE): broadest LoRaWAN range: Dragino, RAK, Milesight, Kerlink, Decentlab, Sensoterra.
• Reichelt.de (DE): MikroTik, Milesight, RAK gateways.
• OpenSprinklerShop.de (DE): Truebner SMT100/SMT50 and a ready LoRaWAN SMT100 node from €99.

UK shops (Wildcare, NHBS, Connected Things) add import VAT and duty for EU buyers since Brexit.

Not verified, open

Bugg v4 pricing (availability resolved 2026-09-13: GroupGets pre-order, first half of 2026, hardware only); Swift successor "Magpie" timing; DIOPSIS price; Sensoterra's two conflicting subscription prices; Milesight's soil-type presets; independent multi-year corrosion reports for the budget probes. If you know, edit this page.