NEWThe Runbaze platform is ready for early adoptersRead the announcement
Platform
Solutions
Company
Who we are
About Careers Contact
Follow along
LinkedIn ↗ Substack ↗
Newsroom
Get in touch Book a demo

Letting AI run the building.

JVJochem VerheulMay 2026 · Case study

Coolnomix delivers IoT middleware that watches energy use across a building: temperature, power draw, room conditions, streamed live from the devices already installed. We built and run the platform behind it. The interesting part is not the dashboard. It is the decision the system takes without asking anyone: turn this down, in this room, now. The moment software adjusts physical infrastructure on its own, the question stops being whether it can and becomes on what grounds.

Because an autonomous decision is only as good as the data underneath it. Sensors drift. Devices misreport. A reading arrives late, or twice, or not at all. A fixed rule, cool when the room passes some threshold, executes bad data with exactly the same confidence as good data. In a dashboard that produces a wrong chart. In a building it produces a wrong temperature, a spike on the bill, or a server room that quietly warms up. Automation without validation is just faster mistakes.

So validation comes before action. Every power reading and device record that enters the platform is checked by models that have learned what plausible looks like for that device, in that room, at that time. Optimisation runs against each room's own baseline rather than a threshold somebody guessed at commissioning. And the work split is deliberate: the machine took the repetitive volume, collecting, validating, adjusting, while the technical team kept judgment, commissioning, exceptions, and the decisions the system is not entitled to make.

01 · SENSE TEMP · POWER ROOM CONDITIONS LIVE · PER DEVICE THE INSTALLED ESTATE 02 · VALIDATE EVERY READING VS LEARNED PLAUSIBILITY BEFORE ANY ACTION 03 · OPTIMISE PER ROOM BASELINE NOT A GUESSED THRESHOLD 04 · ADJUST SETTINGS CHANGED DECISION KEPT EXCEPTIONS → PEOPLE COMFORT HELD · WASTE CUT AUTONOMY OVER PHYSICAL SYSTEMS · VALIDATION BEFORE ACTION
The loop · nothing acts on a reading the system cannot vouch for

The numbers we will put our name to: onboarding a site, once weeks of manual data collection and configuration, now happens in hours. The repetitive load on the technical team fell away with it. And the same architecture runs a single office or a multi-site estate without additional configuration. The number we will not print is a single energy-savings percentage. Every building starts from a different baseline, and the honest figure is the one measured against yours, not an average flattering enough to publish.

Step back and the shape is the one we install everywhere: a model that proposes, an independent check that validates, people holding the decisions with consequence, and a record of what was adjusted and why. In software the consequence is a bad merge. In a building it is physical: comfort, cost, emissions. The higher the consequence, the more the validation layer is the product, not an accessory to it.

Coolnomix now runs as a self-optimising energy platform: live device data in, validated decisions out, at a scale from one office to a global estate. If you want that shape on your own operation, agents on the volume, validation before action, a person at the gate, the sequence starts small and proves itself first.

Start with the Agent Case How people stay in charge← All posts