A beehive is one of the most communicative systems in nature — it just doesn't speak in words. It speaks in the pitch of forty thousand wingbeats, in vibrations that ripple through the comb, and in the shifting cocktail of scents its bees release. Beekeepers have read these cues by ear and nose for centuries. The job of a monitor is to read them continuously, and the job of the AI is to know what they mean.
Channel oneSound: the colony's voice
A healthy colony hums with a fundamental tone around 200–350 Hz — the blended sound of workers fanning, moving, and going about their work. That baseline hum rises and falls with population and activity. But the diagnostic gold is in the specific frequencies, because bees make deliberate sounds at deliberate pitches. ApisNode splits the microphone into ten frequency bands so those signals can be tracked separately.
The acoustic signatures ApisNode listens for
~200 Hz
Worker piping — pre-swarm
200–260 Hz
Swarming-indicator band
260–350 Hz
Colony fundamental — queen-right
300–380 Hz
Queen quacking — in the cell
400–500 Hz
Queen tooting — emerged virgin
500–1000 Hz
Full-spectrum swarm excitation
1–3.6 kHz
Broadband distress energy — hissing's lower edge
Read that way, a spectrum tells a story. Energy climbing in the 200–260 Hz and 420–500 Hz bands over days is a classic pre-swarm signature — the colony broadcasting its intentions before it leaves. A virgin queen tooting at 400–500 Hz during a recovery says a new queen has emerged. And a colony that goes quiet and loses its structured peaks over 24–48 hours may have lost its queen. None of these show up in a single temperature reading; all of them show up in the sound.
Channel twoVibration: the comb as a drum
Bees also communicate through the comb itself, in substrate-borne vibrations an accelerometer picks up directly — and unlike a microphone, an accelerometer barely cares about propolis or wind noise. ApisNode reads six vibration bands across the sensor's resolvable range (5–200 Hz). The most legible is the 160–200 Hz band: wing-fanning — the behavior bees use to regulate temperature and cure honey — beats at a fundamental near 174 Hz, with the colony's collective ventilation hum settling around 190 Hz. When we average vibration by the hour, that band — and every band with it — rises through the day and falls at night, the mechanical heartbeat of a colony working on a daily clock. Down at the very bottom, the 5–20 Hz band carries the "prepare for activity" vibration signals that published research links to swarm preparation weeks in advance.
A note on physics
The accelerometer samples at 400 Hz, so it can only resolve vibrations up to 200 Hz. Higher-frequency behaviors — the waggle dance (200–300 Hz) and queen presence (300–500 Hz) — live above that limit and are read from the audio spectrum instead. Using each sensor only where it's physically valid is part of reading a hive honestly.
Channel threeGas: the smell of change
The third channel is chemical. A metal-oxide gas sensor responds to the volatile organic compounds (VOCs) in hive air — the alcohols and esters of curing nectar, the terpenes of fresh propolis, the pheromones of expanding brood. ApisNode doesn't take one gas reading; it sweeps the sensor's heater through ten temperatures (100–400 °C) each cycle, because different compounds react most strongly at different plate temperatures. The result is a ten-point fingerprint of the air, not a single number — and a change in the shape of that fingerprint signals a change in the kind of VOCs present.
Crucially, the gas sensor can't name a compound or diagnose a disease — no metal-oxide sensor can. It answers a narrower, honest question: has the air changed, and does that change line up with anything else the colony is doing?
The synthesisFrom bands to a report
Raw bands aren't an answer; they're evidence. Turning evidence into a health assessment is where the AI comes in — and where it has to be kept honest.
-
Measure & compress on the hive
The node runs the spectral analysis on-chip and sends compact band values — 30 metrics every five minutes — rather than raw audio, which keeps the radio link tiny.
-
Fuse 24 hours of context
The cloud assembles a day of every signal per colony, plus the outdoor sensor baseline and any beekeeper inspection notes — so no metric is judged alone.
-
Ground the model in the literature
Before it reasons, the AI retrieves relevant findings from a curated library of peer-reviewed bee science (the frequency thresholds above come straight from it). This retrieval-augmented grounding ties conclusions to published evidence instead of a model's guesswork.
-
Compare to the colony's own baseline — and its own hive
Every judgment is relative to this hive's history — because a "normal" gas resistance or sound level varies hive to hive. Deviations, not absolutes, are what matter. And when the keeper has described their equipment with the Hive Configuration tool, the analysis also knows where the sensor sits and what the hive is built of — a peripheral sensor in an insulated Apimaye box is graded very differently from a brood-center sensor in 19 mm pine.
-
Write the report
The model produces a plain-language daily assessment: a health score, the signals behind it, and specific recommended actions — the same report our monitored beekeepers read each morning.
Why grounding matters
An ungrounded language model will confidently invent a plausible-sounding diagnosis. Grounding it in retrieved, peer-reviewed findings — and forcing every call to be relative to the colony's own baseline — is what turns "AI hive analysis" from a gimmick into something a beekeeper can trust. We even show our work: the story below points out where our own field data reproduces the published science the analysis is built on.
Show the workThe bibliography behind the analysis
"Grounded in peer-reviewed science" is easy to say, so here is the actual reference library the analysis retrieves from — 200 sources spanning thermoregulation, acoustics, vibration, gas/VOC sensing, pests and pathogens, treatments, environmental stressors, and colony-health modeling, from Woods (1959) and Seeley's foundational colony work through 2026 machine-learning and thermal-physics papers. It grows as the literature does; this list is generated directly from the knowledge base the AI actually cites.
Browse all 200 references
- 'A review of short-term weather impacts on honey production' 2024, Int. J. Biometeorol., doi:10.1007/s00484-024-02824-0 (PMC11785677)
- Abdollahi et al. 2026 (arXiv:2607.20386) + UrBAN (Sci. Data 2025)
- Abidin et al. 2022, Sensors 22:3301
- Alburaki & Corona 2021/22, J. Apic. Res
- Analog Devices 2021, 'Elusive Tones: Aliasing Effects in Digital MEMS Accelerometers in Condition Monitoring.'
- Annals Entomol. Soc. Am. 2021, 114(5):596
- Arias-Calluari et al. 2026, J. R. Soc. Interface 23:20250505
- Asiri et al. 2026, Ecol. Evol. 16:e73768
- B-GOOD Consortium 2024
- Bankova et al. 2000, Apidologie 31:3–15
- Beling 1929 (Zeitgedächtnis)
- Benaets et al. 2017, Proc. R. Soc. B 284:20162149
- Bencsik et al. 2011, Comput. Electron. Agric. 76:44
- Bencsik et al. 2023, Sensors (PMC10099037)
- Bencsik et al. 2023, Sensors 23:3588
- Bikaun et al. 2022, Sci. Total Environ. 845:157123
- bioRxiv 2025 M. plutonius surveillance (preprint — flag)
- bioRxiv 2025 railway vibration (preprint)
- Boch & Shearer 1966, J. Apic. Res. 5:65–70
- Boch, Shearer & Stone 1962, Nature 195:1018–1020
- Bosch Sensortec BME688 datasheet & BME AI-Studio gas-classification workflow
- Bricout et al. 2024, Sensors 24(18):6067 (Bee Together)
- Burgues & Marco 2018, Sensors 18(2):339
- Bąk et al. 2022, Sensors 22:1148
- Capaldi et al. 2000, Nature 403:537–540
- Capela et al. 2022, STOTEN (sulfoxaflor homing)
- CEA 205:107589 (2023 review)
- Cepero et al. 2023
- Cho et al. 2021 (PMC7882926)
- Clarke & Robert 2018, Apidologie 49(3):386–396, doi:10.1007/s13592-018-0565-3
- Coallier et al. 2025, Commun. Earth Environ., doi:10.1038/s43247-025-02082-x
- Colin et al. 2025, Ecol. Inform. (single-sensor EWS)
- COLOSS B-RAP 2024, Bee World (LLM evaluation)
- Cook et al. 2022, J. Econ. Entomol. 115:715
- Cook et al. 2025 (PMC12219277)
- 'Correction Model for Metal Oxide Sensor Drift Caused by Ambient Temperature and Humidity,' Sensors 2022, 22(9):3301
- E-β-ocimene / brood-cannibalism 2024/25 (PMC11801708)
- EBV/EBV+ (arXiv:2402.01902, ACM TKDD 2025)
- Ecological Indicators 2024 (S1470160X24014183)
- Ecological Indicators 2024 (S1470160X24014183, overwintering brood status / mortality)
- Ellis & Delaplane 2008
- Ellis et al. 2008, J. Insect Physiol. 54:1516–1521
- Ellis, Nicolson, Crewe & Dietemann 2010, Naturwissenschaften 97:429–433
- Eouzan, Garnery, Pinto et al. 2019 (PLOS ONE / PMC6368279)
- EPILOBEE (PMC5344352)
- Fahrenholz, Lamprecht & Schricker 1989, J. Comp. Physiol. B 159:551–560
- Ferrari et al. 2008
- Ferrari et al. 2008, CEA 64:72
- Ferrari R.M. et al. 2025, Commun. Earth Environ
- Ferrari R.M. et al. 2025, Commun. Earth Environ. (air quality)
- Fletcher 1975, Nature 256:721–723
- Flores et al. 1996, Apidologie 27:185
- Free 1958, J. Agric. Sci. 51(3):294–306
- Frizzera et al. 2020
- Genersch et al. 2010, Apidologie 41:332
- Giuffre et al. 2023
- Gomez et al. 2025
- Graham et al. 2006
- Hall, Bencsik & Newton 2023, Sci. Rep. 13:10202
- Harrison et al. 2026, J. Exp. Biol. 229:jeb252357 (PMC13450896)
- Harrison et al. 2026, JEB
- Hayes et al. 2019, Entomol. Exp. Appl. 167 (SHB slime volatiles)
- He, Tian, Wu & Zeng 2016, Insect Science 23(1):157–159, doi:10.1111/1744-7917.12298
- Heliyon 2024 (mixture potentiation)
- Hepburn & Radloff 1998, Honeybees of Africa (Springer)
- Hrncir et al. 2011
- Huet et al. 2026, Sci. Rep. 16:7080 (BeeViz)
- Human et al. 2006
- Human, Brodschneider, Dietemann et al. 2013 (COLOSS BEEBOOK), J. Apic. Res. 52(4)
- Human, Nicolson & Dietemann 2006
- Human, Nicolson & Dietemann 2006, Naturwissenschaften 93:397–401
- Hunt lab 2017, Behav. Ecol. Sociobiol. (comb damping)
- Insectes Sociaux 2025 ('Honey bee beards')
- Insects 17:440 (2026, thymol/OA markers)
- Inventions 10:23 (2025, robbing IoT)
- J. Apic. Res. 2017 (formic 60% protocol)
- J. Econ. Entomol. 2021 (diutinus storage)
- J. Insect Physiol. 2023 (varroa × cold)
- Jack et al. 2023, Front. Ecol. Evol. 11:1102457
- Jack, van Santen & Ellis 2020, J. Econ. Entomol. 113:582
- Jay 1965–1968, J. Apic. Res. 4–7
- JIPM 2022 SHB review
- Johnson 2015, Annu. Rev. Entomol. 60:415–434
- Jones et al. 2004
- Jones, Myerscough, Graham & Oldroyd 2004, Science 305:402–404
- Kadlikova et al. 2022, Agriculture 12:812
- Kanelis et al. 2023, 'Decoding the Behavior of a Queenless Colony Using Sound Signals,' Biology (Basel) 12(11):1392, doi:10.3390/biology12111392 (PMC10669568) [journal is Biology, not Insects]
- Kanelis et al. 2023, Biology 12:1392
- Kleinhenz, Bujok, Fuchs & Tautz 2003, J. Exp. Biol. 206:4217–4231
- Książek et al. 2025, Sensors 25(14):4424
- Kulyukin et al. 2018, Appl. Sci. 8:1573
- Kulyukin et al. 2021
- Kwadha et al. 2017, Insects 8:61
- Lee et al. 2020, J. Insect Physiol. 122:104040
- Lundie 1925
- Maisonnasse et al. 2009, Apidologie 40:562–564
- Maisonnasse et al. 2010, PLoS ONE 5(10):e13531
- manufacturer specs surveyed 2026-09 (docs/HIVE_CONFIG_SPEC.md)
- Mattila et al. 2021, R. Soc. Open Sci. 8:211215
- Meikle & Holst 2015, Apidologie 46:10–22
- Meikle & Weiss 2025 (PMC12219277)
- Meikle & Weiss 2025, Sci. Rep. 15:22042
- Meikle et al. 2015, Apidologie 47:1–14
- Meikle et al. 2016 (PLoS ONE, imidacloprid), 2019 (methoxyfenozide), 2021 (Sci. Rep., clothianidin)
- Meikle et al. 2017, Apidologie 48:666–680
- Meikle et al. 2022 (HAL hal-04183757)
- Michelsen et al. 1986, J. Comp. Physiol. A 158:605–611
- Michelsen et al. 1992, Behav. Ecol. Sociobiol. 30:143–150
- Michelsen, Kirchner, Andersen & Lindauer 1986, J. Comp. Physiol. A 158:605–611
- Minaud et al. 2024, Ecol. Indic. 169:112961
- Minaud et al. 2024, Ecological Indicators 169 (S1470160X24014183)
- Mitchell 2016, Int. J. Biometeorol. 60:629–638
- Mitchell 2017, Interface Focus (hive humidity)
- Mitchell 2023, J. R. Soc. Interface 20:20230488
- Molecules 2021, 26(11):3462
- Molina-Montenegro et al. 2023, Sci. Adv. (HV towers)
- Moore et al. 1989, J. Biol. Rhythms 4(3):305
- Newton et al. 2024, Appl. Sci. 14:1679
- Nieh & Tautz 2000
- Nieh & Tautz 2000, J. Exp. Biol. 203:1573–1579
- Nieh 1993, Behav. Ecol. Sociobiol. 33:51–56
- Nolasco & Benetos 2018 (DCASE)
- Nolasco et al. 2019
- Nolasco et al. 2020
- Nyquist–Shannon sampling theorem (standard DSP)
- Nürnberger et al. 2018 PeerJ 6:e4801, 2019 Oecologia 189:1121
- Ormeño-Arriagada et al. 2026, Insects 17(6):547
- Owens 1971, USDA Tech. Bull. 1429
- Palmer-Young et al. 2023, Commun. Biol. 6:333
- Papachristoforou et al. 2008, Apidologie 39:468
- Pellegrino et al. 2013, PLoS ONE 8(10):e75004, doi:10.1371/journal.pone.0075004
- Peters, Gravish & Combes 2017, J. Exp. Biol. 220:2203–2209
- Peters, Peleg & Mahadevan 2019, J. R. Soc. Interface 16:20180561
- Pettis & Shimanuki 1999
- PLoS ONE 2025 (E-β-ocimene cannibalism)
- PMC11627422 (2024 seasonal physiology review)
- PMC11766442 (2025 urban vitellogenin)
- Qandour et al. 2014
- Rajagopalan et al. 2024, Sci. Rep. 14:6410
- Ramsey et al. 2018, Sci. Rep. 8:14571 (DVAV)
- Ramsey et al. 2020
- Ramsey, Bencsik & Newton 2017, PLOS ONE 12(2):e0171162
- Ramsey, Bencsik et al. 2018, Sci. Rep. 8:14571
- Ramsey, Bencsik et al. 2020, Sci. Rep. 10:9798
- Ramsey, Bencsik, Newton et al. 2020, 'The prediction of swarming in honeybee colonies using vibrational spectra,' Sci. Rep. 10:9798
- Ramsey, Bencsik, Newton et al. 2020, Sci. Rep. 10:9798
- Rangel & Seeley 2008, Anim. Behav. 76(6):1943–1952
- Rittschof & Seeley 2008, Anim. Behav. 75:189–197
- Rojas-Nossa et al. 2022, Pest Manag. Sci. 78:5142
- Roy & Beer 2025, Curr. Opin. Insect Sci. 68:101327
- Seeley & Tautz 2001
- Seeley & Tautz 2001, J. Comp. Physiol. A 187:667–676
- Seeley 1974
- Seeley 1974 (CO2 levels)
- Seeley 1974, J. Insect Physiol. 20:2301
- Seeley 1974, J. Insect Physiol. 20:2301–2305
- Seeley 1985 & 1995 (colony biology, thermoregulation)
- Seeley 1985, Honeybee Ecology (Princeton UP)
- Seeley 2010, Honeybee Democracy (Princeton UP)
- Senger et al. 2023, Smart Agric. Technol. 6:100329
- Southwick & Moritz 1987, Int. J. Biometeorol. 31:259–265, doi:10.1007/BF02188929
- Southwick & Moritz 1987, J. Insect Physiol. 33:623–626
- Southwick 1983
- Southwick 1983, Comp. Biochem. Physiol. 75A:641–645
- Southwick 1985, J. Comp. Physiol. B
- Southwick 1991, Bees and Beekeeping
- St. Clair, Beach & Dolezal 2022, PLoS ONE 17:e0266219
- Stabentheiner et al. 2003, J. Exp. Biol. 206:353–358
- Stabentheiner, Kovac & Brodschneider 2010, PLoS ONE 5(1):e8967
- Stabentheiner, Pressl, Papst, Hrassnigg & Crailsheim 2003, J. Exp. Biol. 206:353–358
- Stalidzans & Berzonis 2013, Comput. Electron. Agric. 90:1–6
- Stalidzans et al. 2017
- Starks & Gilley 1999, Naturwissenschaften 86:438–440
- Swanson et al. 2009, J. Chem. Ecol. 35:1108
- Szczurek & Maciejewska 2020–2021 (Sensors 20:2532, 21:4019)
- Szczurek et al. 2020, Sensors 20:117/2532
- Tautz 1996
- Tautz 2008, The Buzz about Bees (Springer)
- Thomson et al. 2023, Animals 13:154 (AFB dogs)
- Torto et al. (fermenting pollen-dough volatiles)
- Treder et al. 2023, STOTEN (RF-EMF)
- UF/IFAS Extension ENY-162
- Underwood & Currie 2003
- van Dooremalen et al. 2024 (B-GOOD Project)
- van Dooremalen et al. 2024, Insects 15:76 (B-GOOD)
- van Nerum & Buelens 1997, Comp. Biochem. Physiol. 117A:445–455
- Walsh et al. 2026, PLoS ONE
- Walsh et al. 2026, PLoS ONE (hangry bees)
- Wang et al. 2016, PLoS ONE 11:e0154547 (cold brood)
- Watmough & Camazine 1995, J. Theor. Biol. 176:391–402
- White 1978, Advances in Food Research 24:287–374
- Winston 1987 (The Biology of the Honey
- Winston 1987, The Biology of the Honey Bee
- Winston 1987, The Biology of the Honey Bee (Harvard UP)
- Woods 1959
- Woyke 1964, J. Apic. Res. 3:17–23
- Yıldız & Karabağ 2026, Exp. Appl. Acarol. 97:23
- Zacepins et al. 2016, Biosyst. Eng. 148:76
- Zaman & Dorin 2023, CEA 210:107906 (OIP)
- Zgank 2021
Auto-generated from the production knowledge base, September 2026. Citations appear as stored (journal-abbreviated); where a report cites a finding, it names its source from this list.
See the signals in real data
Our 90-day data story renders the actual acoustic, vibration, and gas spectrograms from a live colony — and shows the analysis reproducing published bee science.
The signals at a glance
| Signal | Frequency / channel | What it can indicate |
|---|
| Colony hum | 200–350 Hz | Baseline activity & population; a queen-right, organized colony |
| Worker piping | ~200 Hz | Rises immediately before a swarm lifts off |
| Queen quacking | 300–380 Hz | A virgin queen still in her cell, signaling readiness |
| Queen tooting | 400–500 Hz | An emerged virgin queen — often during post-swarm recovery |
| Waggle dance | 200–300 Hz (audio) | Active foraging communication |
| Wing-fanning | 80–120 Hz (vibration) | Thermoregulation & honey curing; tracks the daily activity cycle |
| VOC fingerprint | gas, 10-step scan | Nectar flow, propolis, brood volatiles, or an environmental air event |
| Hissing | 300 Hz–3.6 kHz | Distress or defensive response |
Keep reading