brings self-learning predictive control to your product portfolio

Your company offers smart thermostats or building automation solutions.

However, you are missing the technology to offer self-learning predictive control to your customers and you want to be ready for future technological trends such as demand response and time-varying energy prices.

We make our control algorithm available to you via an intuitive API.

We provide Predictive control as a Service: Your devices send measurements of room temperatures and actuator states to our cloud, and our algorithm makes recommendations for your future control inputs. Alternatively, our software can be run in your cloud environment directly.

Our solution is


By using predictive control, we find optimal control inputs while taking into account the future behaviour of the building, user preferences and the weather forecast. This leads to a reduced energy consumption while increasing comfort.


Our algorithms learn the thermal behaviour of the connected building by themselves. They do this faster than other approaches, because we combine physics-based knowledge with Machine Learning methods. 1-2 weeks of training data is usually enough to build a model.


Our approach is lightweight and thus scalable. In contrast to other Machine Learning based approaches, the resulting optimization problem is convex and can be solved fast and for large systems. It is also ready for variable energy pricing and demand response schemes.

Frequently Asked Questions

What is the technology behind your solution?

Our technology builds on a combination of Machine Learning methods and physics-based constraints. The Machine Learning methods allow to capture the building dynamics from measurement data, while the physics-based constraints significantly lower the required amount of training data and ensure physical model behaviour.

Are the methods validated?

Yes. Our control method has been validated in different apartment and office units of the NEST demonstrator building at Empa as well as in a light commercial building in a pilot project with the market leader for thermostatic valves. In all cases it saved 20-40% of energy while increasing comfort, compared to state-of-industry controllers.

What do you mean by 'ready for demand response'?

By combining Machine Learning with physics-based constraints, we have developed models that are lightweight and informative at the same time. The resulting convex optimization problem is computationally efficient and can be used efficiently for day-ahead reserves planning or variable energy prices.

Which hardware is required to use your controllers?

Our technology requires one temperature measurement and one actuator state (for example a radiator valve position) per controlled zone. The sensor and actuator data needs to be accessible online. Our solution is therefore perfect for smart thermostats that are already cloud-connected, or cloud-connected building monitoring/automation systems.

Is the energy saving potential the same for every building?

There is savings potential in every building. The amount depends on the type, being the highest for weakly insulated buildings. Besides energy savings, predictive control has several other advantages: increased comfort, efficient use of variable energy prices and readiness for demand-response schemes.