Waking the building

Concept project · fictional company · design study
Check your siteTalk to KELO →
Intelligent buildings · Distributed AI

Make every building intelligent.

KELO connects the equipment buildings already have, adds intelligence at the edge, and turns spare capacity into AI compute.

01 · The problem

Most buildings run blind.

85%+ of U.S. commercial buildingshave no BAS

Equipment runs without anyone seeing how it behaves, until something fails or the energy bill arrives.

02 · See every floor

Sensing on every floor, whatever the brand.

Air qualityOccupancyLeaksPower meteringAI video
03 · Every protocol

Every device speaks a different language.

Wireless sensorsBACnet HVACModbus meters & plant
04 · Universal gateway

Many protocols. One managed edge.

The Universal Building Automation Gateway brings them into the KELO Managed Edge, with local AI inference on the device, so decisions don't wait for the cloud.

Now online

Your building, awake.

Energy prediction, presence, equipment health and automation, managed in KELO Cloud across every site you run.

One platform

Existing systems in. Intelligence out.

No rip-and-replace. KELO bridges the devices a building already has, then runs the applications on top.

Wireless sensorsair, occupancy, leaks
BACnetHVAC, controllers
Modbusmeters, plant
Universal GatewayKELO Managed Edge · on-device inference
KELO Cloudportfolio, models, fleet
Your systemsBMS, dashboards, APIs
Floor 06 · illustrative

Distributed intelligence

Sense. Context. Reason. Act. Verify.

AI that acts on real equipment needs guardrails. Follow one event on one chiller, from first signal to confirmed result.

01 · Sense

A vibration sensor drifts.

Rooftop chiller 2 starts vibrating outside its usual band at 14:06. On its own, it is just one noisy reading.

02 · Context

The gateway lines up the evidence.

Vibration, supply temperature and power draw are aligned by time, location and asset, across three different protocols.

03 · Reason

Edge AI makes a call.

A local model running on the gateway flags likely bearing wear and estimates remaining useful life, without a cloud round trip.

04 · Act

Inside approved limits only.

The workflow lowers the chiller's load to 70% and opens a work order. Rules and permissions decide what AI is allowed to touch.

05 · Verify

Confirm, then report.

Vibration returns to normal, the result is checked, and the enriched event goes to KELO Cloud for the whole portfolio to learn from.

Chiller 2 · Building 014Sense
The next layer · Distributed AI compute

Every building has power it never uses.

Once a building is metered and intelligent, KELO can see something else: electrical capacity that sits idle most of the year.

01 · The electrical service

Sized for a peak it rarely reaches.

The glass is the site's electrical service: 1,000 kW. Nothing can go above it.

02 · Measure, don't guess

120 days. Every half hour.

Service 1,000 kWMeasured peak 680 kW

Each ridge is one day of real demand. Weekends dip, summer climbs, and the highest point is the measured peak.

03 · Headroom

320 kW nobody is using.

The gap between the highest day and the service limit is where new load can live safely.

04 · Compute where the power is

Four Pod60s. Inside the existing service.

each slab = 1 Pod60 · 60 kWup to 128 GPUs
05 · Combined peak

Building plus compute. Still under the glass.

KELO Cloud coordinates building load and AI compute together, so the combined peak stays inside the site's electrical service.

AI-ready asset

Your building, now part of the AI economy.

Lower operating costs and new value from the same power, managed from one cloud.

AI compute · Pod60 + CoolPod

A micro data center that fits where the power is.

A ~60 kW modular system, deployed and operated by KELO at qualified sites.

Payload

Up to four 8-GPU servers. Up to 32 GPUs.

Direct liquid cooling at the chip: cool supply in, heat carried straight out.

CoolPod

Heat leaves. Efficiency stays.

The CoolPod rejects roughly 205,000 BTU/h, while KELO Cloud watches server health, power and site conditions around the clock.

Built to refresh

New GPUs, same interface.

As GPU generations change, servers swap through the same power and cooling connection.

~60 kWplanning power envelope
32 GPUsup to four 8-GPU servers
415 / 480 Vthree-phase service
2027pilot deployments targeted

Illustrative figures. Planning envelopes, not specifications.

Site power check

How much compute can your site hold?

Move the sliders to your electrical service and measured peak. We keep a safety margin and match it to KELO's compute family.

Headroom320 kW
Reserve kept (15%)48 kW
Usable for compute272 kW
Suggested4 × Pod60

Illustrative estimate only. A real site power review also checks voltage, panel space, cooling and combined peak.

How deployment works

From site review to remote operation.

KELO deploys and operates every system. On-site work is scheduled only when it is needed.

Step 1

Site review

Electrical service, space, cooling, connectivity and access.

Step 2

Site engineering

Feeder design, heat rejection and approvals with the engineer of record.

Step 3

Install & commission

Compute, gateways, meters and sensors, verified before handover.

Step 4

Remote operation

KELO Cloud watches server health, power and site conditions.

Step 5

Hardware refresh

New GPU generations through the same power and cooling interface.

One platform

Intelligence at every site. Management across the network.

Environmental intelligencetemperature, humidity, air quality
Power & energysite power, usage, performance
Security & site awarenessAI video, access, occupancy
Remote fleet managementvisibility, alerts, operations
KELO Cloud · portfolio48 sites · 0 alerts
Get started

Let's talk about your site, portfolio or AI infrastructure.