ApisNode is a production-deployed smart beehive monitor built by ApisNode Inc. — 30 real-time metrics per hive every five minutes (environment, sound, vibration, and a gas/VOC fingerprint), daily AI colony-health reports grounded in peer-reviewed bee science, and over-the-air firmware updates. No WiFi at the hive, no manual data collection, no interrupting the bees.
🌐 Live at apisnode.com — live telemetry from production hives, an AI beekeeping assistant grounded in real sensor data, and a public archive of daily colony reports.
🐝 Open Apiary Project — our proposed 501(c)(3) non-profit placing free ApisNode monitors with beekeepers to build an open, anonymized apiary-health network.
🎙️ Heard on Beekeeping Today Podcast — Episode 402.
About this repository. ApisNode is closed-source. This repository is a technical overview for beekeepers, researchers, and partners — what the system senses, how the pieces fit, and where it runs. It is not a build specification: firmware, signal processing, radio protocol, cloud code, analysis prompts, and the science knowledge base are proprietary to ApisNode Inc. See License.
A battery-powered sensor node sits inside the hive. Every cycle it listens (microphone), feels (accelerometer), and smells (gas sensor), alongside temperature, humidity, barometric pressure, and tilt. Spectral features are computed on the node itself, so only a compact packet of band values leaves the hive — never raw audio. The packet travels over encrypted sub-GHz LoRa radio to an always-on gateway, which relays it to the cloud.
Each night, an AI analysis reads the last 24 hours of every monitored colony against its own history, the gateway's outdoor baseline, the keeper's inspection notes, and a curated library of bee-science literature, then writes a plain-language report. The reports put real evidence behind the decisions the beekeeper already makes — the system never tells anyone how to keep bees.
┌──────────────────────────────────────────────────────────────┐
│ APIARY (edge hardware) │
│ │
│ ┌─────────────────┐ encrypted ┌──────────────────┐ │
│ │ ApisNode │ sub-GHz LoRa │ ApisMind │ │
│ │ in-hive node │ ───────────────► │ gateway │ │
│ │ battery, sleeps│ ◄── config / OTA │ mains-powered │ │
│ └─────────────────┘ └────────┬─────────┘ │
│ one gateway serves a whole apiary │ │
└─────────────────────────────────────────────────┼─────────────┘
│ WiFi · MQTT · TLS
┌──────────▼───────────┐
│ AWS cloud │
│ ingest · storage · │
│ API · nightly AI │
└──────────┬───────────┘
┌────────────────────────────┼────────────────────────────┐
│ │ │
┌────────▼────────┐ ┌──────────▼──────────┐ ┌──────────▼──────────┐
│ Mobile app │ │ Wall display │ │ Fleet dashboard │
│ iOS / Android │ │ display.apisnode.com │ │ fleet.apisnode.com │
└─────────────────┘ └─────────────────────┘ └─────────────────────┘
Two complementary ESP32-C6 devices work together at each apiary.
Battery powered (solar-topped in the field), installed between frames. It spends almost all of its time in deep sleep and wakes on a sense → process → transmit → sleep cycle.
| Channel | Sensor | What it captures |
|---|---|---|
| Environment | Bosch BME688 | Temperature, humidity, barometric pressure |
| Smell | Bosch BME688 gas sensor | A multi-step heater sweep that yields a volatile-organic-compound fingerprint, not a single gas number |
| Feel | ADXL362 accelerometer | A six-band vibration spectrum of the hive body, plus tilt (which also tells the cloud when a node has been pulled from the hive) |
| Hear | PDM MEMS microphone | A ten-band acoustic spectrum and overall colony sound level |
Mains powered, installed outside the hives. Runs a continuous LoRa receiver so nodes never have to synchronize with it, bridges the apiary radio network to the cloud over WiFi, and provides an outdoor environmental baseline for the analysis. It also handles BLE provisioning from the mobile app and store-and-forward buffering when the uplink drops. One gateway serves an entire apiary.
🔊 Sound. The node resolves ten acoustic bands placed over the frequency ranges where honey bee colonies make deliberate, documented sounds — the baseline colony hum, worker piping ahead of a swarm, queen quacking and tooting, and the broadband hiss of a defensive colony. Tracking each band against its own history is what lets the analysis see a signature build over days rather than react to a single noisy reading.
📳 Vibration. Six vibration bands inside the accelerometer's resolvable range capture the mechanical side of colony life: wing-fanning for thermoregulation and honey curing, the daily rise-and-fall of activity, and low-frequency signals that published research associates with swarm preparation. Unlike a microphone, the accelerometer is largely immune to propolis and wind.
🌬️ Smell. Instead of one gas reading, the sensor's heater is swept through a series of temperatures each cycle, because different compounds react most strongly at different plate temperatures. The resulting profile is compared with the colony's own baseline to flag changes in hive air chemistry — nectar curing, propolis, brood volatiles, or an environmental air event worth a look at the next inspection.
🌡️ Environment. Temperature, humidity, and pressure inside the hive, read with the gas heater off so they stay clean, plus the gateway's outdoor reading as a hyper-local weather reference.
The exact band plan, heater profile, and on-node DSP are part of the proprietary firmware. The public explainer How AI reads a beehive describes the signals in beekeeper terms.
A nightly serverless job fuses several sources for each colony and asks a Claude model on Amazon Bedrock to write the report:
- Bee-science knowledge base — a curated, citation-backed library of peer-reviewed findings organized by sensor modality (acoustics, vibration, air quality, temperature, seasonal behavior).
- 24 hours of telemetry — all 30 metrics, read against the colony's own multi-week history, plus the gateway's outdoor baseline cross-checked against a weather service.
- Hive configuration — hive type, insulation, and setup entered by the keeper, so expectations match the box the bees are actually in.
- Inspection notes — voice-transcribed field observations with AI-extracted labels.
- Prior reports — what was flagged before, so the narrative has continuity.
Output per colony: a status (thriving · normal · watch · attention), the evidence behind it, and any observation worth checking at the next inspection — delivered in the app, on the wall display, and (for participating apiaries) in the public report archive.
The analysis is deliberately conservative. It follows a minimal-intervention, phenomenology-first philosophy: describe what the sensors show, cite the science that explains it, and leave the husbandry decision to the beekeeper. It is also careful about its own blind spots — it knows when a gateway is indoors, when a node has been pulled from the hive, when a cool-down is explained by the weather, and when a sensor's reading should not be compared with an older generation of firmware.
An interactive assistant extends the same grounding to chat: it can pull the full telemetry history of your hives, draw charts inline, and export CSV on request.
- Sub-GHz LoRa, node → gateway. Every packet is encrypted and authenticated, so telemetry cannot be read off the air and configuration or firmware commands cannot be forged. Typical range is around two miles line-of-sight; details of the link are on Beehive monitoring without WiFi.
- Acknowledged delivery. Readings are acknowledged by the gateway, de-duplicated, and retried on a gated schedule with transmit-time jitter, so colonies sharing one gateway do not stack on each other. The fleet dashboard grades true delivery per node from the packet sequence.
- Zero-touch pairing. A new node pairs with its gateway on first boot when held close to it; keys are exchanged only in that proximity window. No configuration screens.
- Delta OTA updates. Both devices take binary delta firmware updates — the gateway over WiFi, the node relayed over LoRa — with integrity checks and automatic rollback if the new image does not validate.
- IoT Core — MQTT ingest of telemetry from ApisMind gateways
- DynamoDB — telemetry store with retention lifecycle and time-bucketed history
- Lambda + API Gateway — REST API, nightly analysis, inspection labeling, OTA and fleet management, store and account services
- Amazon Bedrock — Claude models for reports, chat, and labeling (Sonnet-class for analysis and chat, Haiku-class for labeling and the public demo)
- Cognito — authentication with server-side device-ownership enforcement; in-app account deletion
- S3 + CloudFront — firmware images and delta patches, inspection audio and transcripts, the public report archive
- Amplify — hosting for the web properties
BLE device provisioning, an animated 30-metric dashboard, live spectrograms, daily AI reports with interactive chat, voice-to-label inspections, hive configuration, telemetry CSV export, team apiaries with member roles, read-only apiary sharing, remote device management (reboot, recalibration, OTA), and an opt-in to the Open Apiary Project showcase.
🖥️ Wall display — display.apisnode.com
Always-on kiosk dashboard for a wall-mounted screen: live graphs with hover crosshairs, 2-D spectrograms on real time axes, battery history, 1 h–30 d ranges, the nightly report, and ApisMind chat. Includes a screen-reader view of the current spectral scan.
📊 Fleet dashboard — fleet.apisnode.com
Operations console: firmware tracking and OTA rollout, hardware statistics (battery, RSSI, check-in and delivery rates, gateway diagnostics), end-of-line test and kit staging, and the trial-application pipeline for the Open Apiary Project.
🌐 Website — apisnode.com
Live telemetry from production hives, the AI assistant, the public daily-report archive, and explainer pages: How AI reads a beehive, Beehive monitoring without WiFi, a buyer's comparison, and 90 days inside a beehive.
Open Apiary Project — openapiaryproject.org
The Open Apiary Project is our proposed 501(c)(3) non-profit (formation in progress). Its mission is to place free ApisNode kits with beekeepers and build the most comprehensive anonymized apiary-health network available to researchers — for colony-loss research, pollination forecasting, and regional disease detection.
- Live from the network — the site publishes daily network-level aggregates: colonies reporting, diurnal activity, thermal-state cohorts, and sound, vibration, and gas trends, with an AI trends summary.
- Where it runs — Northern New Mexico (the founding research apiary, with public daily reports), the Florida Gulf region, the Montana Rockies, and Atlantic Nova Scotia. Locations are shown at region level only; exact apiary locations are never collected or published.
- Privacy and consent — participation in the public showcase is opt-in per apiary; individual colony reports stay private to the keeper unless they choose otherwise.
- Research access — anonymized data bundles are available to researchers under a data-use agreement.
US beekeepers lose on the order of 40–50 % of managed colonies a year. Continuous, non-invasive monitoring is one of the few tools that can see trouble building before an inspection would.
Learn more and apply → openapiaryproject.org
- July — science knowledge base expanded to 68 verified literature findings; minimal-intervention analysis philosophy with per-band spectral rhythm analysis; AI chat gained full-history telemetry access and inline charts.
- August — acknowledged LoRa delivery with de-duplication and jitter; fleet hardware-statistics tab with sequence-based delivery grading; the network's first site outside the US (Nova Scotia); Beekeeping Today Podcast episode 402.
- September — Open Apiary Project "Live from the network" page with a lit globe and daily aggregates; telemetry CSV export; research data-access tier; hive-configuration tool with vendor insulation presets; read-only apiary sharing and team apiaries; microphone front-end rework with bench tone-test verification.
- October — out-of-hive node detection; stores-trajectory and cool-season advisories; indoor-gateway guard and weather-service outdoor reference; gateway-thermometer self-heat correction; accelerometer shaker-stand test for end-of-line QA.
| Layer | Technology |
|---|---|
| Firmware | ESP-IDF v5.5, C, FreeRTOS, ESP32-C6 |
| Radio | Seeed Wio-E5 LoRa module, sub-GHz US ISM band |
| Sensors | Bosch BME688, ADXL362, PDM MEMS microphone |
| DSP | On-node spectral feature extraction (proprietary) |
| Mobile | Flutter 3.x (Dart), iOS & Android |
| Cloud | AWS CDK v2, Lambda, DynamoDB, IoT Core, Bedrock, Cognito, S3, CloudFront, Amplify |
| AI | Claude models on Amazon Bedrock |
| OTA | Binary delta patches with integrity verification and rollback |
| Web | Hand-written HTML/CSS/JS on AWS Amplify |
Proprietary — © 2026 ApisNode Inc. All rights reserved.
The ApisNode system — hardware designs, firmware, signal processing, radio protocol, cloud services, analysis methods and prompts, the bee-science knowledge base, and the applications — is closed-source. No license to reproduce, reverse-engineer, or build derivative systems is granted by this repository, which exists solely as a public technical overview. ApisNode and ApisMind are trademarks of ApisNode Inc.
Inquiries: info@apisnode.com · apisnode.com · openapiaryproject.org