Skip to content
hum

How it works

The first second
says what it is.

Scroll slowly. Below, a compressor starts. The curve comes from the equation for a motor, and next to it you see what hum shows on its screen at each moment.

0 ms

A house at rest.

The draw is steady. Nothing has changed, so there is nothing to explain.

Right now, on the hub
Not identified

Current model: inrush of about four times running current, decaying with a 55 ms time constant, lagging the voltage. Timescale at 50 Hz.

hum recognised the kind of machine.
You gave it the name.

Three layers, kept apart

Measured, inferred and named.

Measured and signed

What went through the wire.

Power, voltage, frequency, energy per hour. This part is certain, and it is what gets sealed into the record.

Inferred and marked

What kind of machine it looked like.

Compressor, heating element, motor, power supply. Physics tells these apart, and the screen marks them as estimates until you confirm.

Named by you

Which machine it actually was.

"Fridge" or "the towel rail in the bathroom". Only someone in the house knows that, so hum asks once and remembers.

Where it thinks

On the device, and it tells you where.

The hub has a screen that states where inference is running. On hum pro that is a neural processor. On a board without one it says "on the processor" and does the same job more slowly.

It compares starts.

A two-second window around each change, including the half second before it, is what gets matched against what the house has already taught it.

It sorts by physics first.

Compressor, resistive, motor, switching. Those four families come out of how the current behaves, before any name is involved.

It shows what it has not learned.

How many appliances it has seen, how many have names, and what share of the house is still unidentified.

Close candidates lower its confidence.

If a start fits two machines almost equally, it offers both and the line stays an estimate.

The hub's Learning screen saying inference runs on the processor because the board has no neural processor, and that no appliance has been learned yet.

The Learning screen on a development board with no neural processor and no sensor attached. It says both.

Two models

hum and hum pro.

Both models keep the same signed record. It is not an extra you pay for.

Featurehumhum pro
Whole-house measurement, every secondYesYes
Signed, chained hourly recordYesYes
Appliance identificationOn the processorOn a neural processor
"Not identified" always shownYesYes
Phone app, approved on the hub's screenYesYes
Per-appliance health: a learned baseline for each machineNoYes

On hum, the health screens say "available on hum pro" and show no empty chart.

See it on your phone.

The app finds the hub by itself, and the hub asks before any phone may connect.

hum is in development; a reservation costs nothing Reserve one