THE SYSTEM
How AirPulse works.
Lower bills, less carbon, comfort you can count on. Here is the machinery underneath that outcome: it learns what each unit's normal looks like, watches every hour against it, and tells you, with a dollar figure, when a unit starts costing you more than it should. You control the temperature from your phone from day one.
01Attach
A palm-sized node connects to each HVAC unit. On Daikin systems it reads the unit's own communication bus. On conventional rooftop units and PTACs it reads the thermostat circuit and sensor probes. A technician installs it in under an hour, or you self-install a mini-split or standard-thermostat unit in 15 to 30 minutes, app-guided, the same class of job as a Nest. No building automation system, no points list, no site engineering.
02Control
From day one you set temperature and setpoint from your phone, in Apple Home or your AirPulse login. The unit shows up in the Home app you already use; ask Siri, build an automation, or adjust a unit across town. Your comfort is in your pocket, and the intelligence runs underneath it.
03Learn
For about 30 days, AirPulse builds a statistical baseline of that specific unit: runtime, temperatures, cycling, compressor behavior, bucketed by outdoor conditions, because normal at 95°F is nothing like normal at 70°F. Every unit gets its own model. The baseline keeps refining across seasons for as long as the unit runs.
04Watch
From then on, AirPulse compares every hour of operation against what that unit should be doing in that weather. Degradation usually shows up in the data before a human feels it: runtime creep, compressed delta-T, cycling changes, current draw rising for the same output.
05Act
When a unit deviates, you and your contractor get a plain-English alert with the likely cause and an estimated monthly cost. “Unit 4's compressor runtime is up 22% for these conditions. Likely dirty condenser coil. Costing about $85 a month” (illustrative; the dollar figure depends on your rate and unit). The alert supports scheduling the fix while it's still small, bringing the second price back down.
AirPulse learns what each unit should be doing right now and tells you when it isn't.
Honesty block
Today's engine is rigorous per-unit statistical baselining: incremental mean and variance per parameter, per outdoor-temperature band, with deviation scoring against the matching band. It's deliberately explainable; you or your engineer can audit the math behind every alert. We do not claim deep learning, and we do not quote failure-prediction accuracy we haven't earned on a large fleet. As the deployed fleet grows, the models grow with it, and we'll say so when they do.
THE SCIENCE
The science is public. Access is still limited.
The steps above describe what AirPulse does. One fair question remains: if the waste is this well documented, why isn't continuous monitoring already on every unit?
Not because the science is secret. Fault detection for HVAC has a deep public research base, built over two decades, and we cite it everywhere we use it. The U.S. Department of Energy says routine filter replacement alone can lower an air conditioner's energy consumption by 5 to 15%. When Downey and Proctor checked 13,258 air conditioners for a 2002 ACEEE Summer Study paper, 57% were outside specification for refrigerant charge. And the canonical review of the field, Katipamula and Brambley (2005), written at Pacific Northwest National Laboratory, estimates that poorly maintained, degraded, and improperly controlled equipment wastes 15 to 30% of the energy used in commercial buildings.
What has stayed limited is access. Much of the tooling built on that research assumes infrastructure: a building automation system, trend logs, a controls contractor, an engineer reading the output. Most commercial buildings under 100,000 square feet run without a building automation system, and the rooftop units, split systems, and PTACs common across New York generally carry no continuous monitoring at all; a maintenance visit might put eyes on a unit for twenty minutes, twice a year. The science exists. The equipment it would help most has largely been outside it.
That gap is what AirPulse exists to close, with subscriptions starting at $29 per unit per month. Making continuous fault detection practical on this equipment took new work on how a single unit's normal is learned and checked. Two U.S. provisional patent applications concerning that fault detection method and per-unit baseline approach are on file with the USPTO. Provisional applications are not examined on the merits and are not issued patents; they record an early filing date while nonprovisional applications are prepared. Selected implementation details are kept confidential.
Every figure above carries its full source on How we make claims.
Watch it learn your equipment.
Launching 2026 · One update a month, no spam