Your hive, right now
The 8 listening bands · 175–525 Hz
The raw acoustic spectrum from the hive mic — eight Goertzel frequency bands. The tallest bar is the colony's dominant tone.
What the shape means
Each band-shape maps onto six possible meanings. The highlighted one is what drove the verdict.
Every number above is a real field from the BeeKeeper engine's /explain endpoint. In DEMO it replays a captured trace; switch to LIVE with the engine running and it polls the colony every few seconds.
How the AI turns a sound into meaning
Eight numbers come off a microphone in a Maine field. Watch them become an interpretation, a verdict, and the cited science that backs it — every value below is a real field from the engine.
1 · Signal
Raw acoustic spectrum — 8 Goertzel bands, 175–525 Hz, straight off the hive mic.
2 · Interpretation
The spectrum's shape maps to six meanings. The one that explains the verdict is highlighted.
3 · Verdict
A frozen classifier — byte-identical to the one running on the Seed — names the state.
4 · Research
The verdict is checked against a vector knowledge base of cited bee-acoustics studies — nearest evidence first.
Two days in the life of Hive #1
A timeline of the colony's state — every reading, every event, scrubbed back hour by hour. This is the dashboard Stuart sees after harvesting the data from the field.
Ask the hive
Everything the system knows about bee acoustics — queenless roars, swarm piping, robbing, thermoregulation — drawn from peer-reviewed studies. Ask a question and hear the sound it describes, synthesized live.
A hive that listens, off the grid
No wifi. No cell signal. No mains power. A solar-charged computer sits in a field in Harpswell, Maine, listening to a beehive and deciding — on its own — what the colony is doing.
The chain of custody
The honesty model
An AI in a field is only trustworthy if it tells you what it cannot do. This one does:
- It states its confidence. Every verdict carries a number. Low confidence is shown as low, never hidden.
- It flags what it cannot hear. The strongest swarm-prediction signal is a ~20 Hz comb vibration — below this mic's floor. The app says so rather than pretending.
- It labels synthetic vs real. When the data source is a simulation, the badge reads synthetic, not live. No theater.
- It shows its work. Every verdict links to the peer-reviewed study behind it. You can check the AI.
Why this is hard (and why it matters)
A beekeeper can open a hive maybe once a week, and only on a warm day. A queen can fail, a colony can prepare to swarm, or robbers can collapse a hive in the days between visits — and by the time you look, it's over. A microphone never blinks. By learning each colony's own baseline and listening for the acoustic signatures the research describes, this system can tell a non-technical keeper something they could otherwise only guess at: your hive sounds calm and queenright — or something changed at 3am, go look. That is AI doing real, useful work in a place with no infrastructure at all.
Built for one hive in Maine, designed to honestly report what it knows and what it doesn't.