The Hardware Shield: why our AI stays advisory

Most companies that promise to make a building smarter want to sell you a box. A new controller, a gateway, an edge appliance, a sensor kit. The box becomes the relationship. It sits on their balance sheet as inventory, it sits in your riser as a dependency, and it quietly decides who you can work with for the next ten years.

puniq is built the other way around. We are an asset-light software company in Berlin. We do not manufacture hardware, we do not resell it, and we do not carry it. We call this the Hardware Shield, and it is not a marketing line. It is the structural choice that keeps our margins clean, our regulatory footprint small, and your building free of lock-in.

The second choice follows from the same instinct. Our AI is advisory only. It analyses, forecasts and flags, but a human approves every action before anything in your building moves. That keeps the intelligence layer out of the heaviest audit regimes by design. Together these two decisions describe the kind of company we are: software, not steel; counsel, not autopilot.

Software, not steel. Counsel, not autopilot. The risk is engineered out on purpose.

What the Hardware Shield actually means

Asset-light is a precise term, not a vibe. It means the value we create lives in software and documentation, not in physical inventory we have to buy, stock, ship, install and warranty. A hardware-heavy integrator carries cost and risk on every project: components on the books, supply-chain exposure, depreciation, RMA logistics, field failures. Each of those is a place where money leaks and timelines slip.

We sidestep all of it. Our three products are software and engineering work:

• Data Auditor unifies an existing multi-vendor building into one readable layer, on-premises, on the controllers you already own.
• Infrastructure Blueprint designs the open automation layer from scratch for new builds and deep retrofits, specifying open protocols rather than proprietary boxes.
• AI Optimization is a cloud subscription layered on top: benchmarking, forecasting, anomaly detection and compliance output.

In none of those do we sell you a panel. When hardware is genuinely needed, you buy it on the open market, in your name, from whatever vendor wins on price and quality. We stay vendor-neutral on purpose. That is the difference between a partner who is paid to make your building perform and a supplier who is paid to keep you buying.

The second shield: advisory-only AI

The same discipline shapes how we use artificial intelligence. It would be easy, and tempting, to let the model close the loop: detect an anomaly, command the valve, override the setpoint, all without a person in the room. Plenty of building-AI vendors pitch exactly that autonomy.

We refuse it on principle and by design. puniq's AI is human-in-the-loop. The system surfaces a recommendation with the evidence behind it. A qualified operator reviews it and decides. The model advises; the human acts. Nothing in your building changes state because an algorithm felt confident at three in the morning.

This is partly engineering judgement. Buildings are safety-critical, occupied, and full of edge cases a model has never seen. An advisory system fails safe: the worst case is a recommendation nobody acts on, not a chiller that trips a hospital wing. But the bigger reason is regulatory, and it is where the two shields connect.

The model advises. The human acts. Nothing moves because an algorithm felt confident at 3am.

Staying outside BSI C5 and NIS2, on purpose

When software takes direct, autonomous control of critical building systems, it walks into the heaviest compliance regimes in Europe. BSI C5, the German cloud-security attestation, and NIS2, the EU directive on essential and important entities, both scale their demands with how critical and how autonomous a system is. Cross the line into controlling essential infrastructure unsupervised, and the audit burden, the liability and the cost climb steeply.

By keeping the AI advisory, with a human approving every consequential action, the intelligence layer stays outside that audit scope by design. It is an assistant to operators, not an autonomous controller of essential services. That is not a loophole we found after the fact. It is the reason the architecture looks the way it does.

For an owner, the payoff is concrete. You get the upside of AI, sharper forecasting, faster anomaly detection, automated compliance reporting for CSRD and the EU Taxonomy, without inheriting the regulatory weight that would come from handing a model the keys. The risk that would otherwise sit on your building stays off it.