The performance of model predictive smart home heating control (SHHC) heavily depends on the accuracy of the initial setup for individual building characteristics. Since owners or renters of residential buildings are predominantly not experts, users’ acceptance of SHHC requires ease of use in the setup and minimal user intervention (e.g. only declaration of preferences), but at the same time high reliability of the initial parameter settings and flexibility to handle different preferences. In contrast, the training time of self-learning SHHC (e.g. based on artificial neural networks) to reach a reliable control status could conflict with the users’ request for comfortable heating from the very beginning. Dealing with this trade-off, this paper follows the tradition of design science research and presents a prototype of an online optimisation tool (OOT) for SHHC. The OOT is multi objective (e.g. minimising lifecycle energy (cost) or carbon emissions) under constraints such as thermal comfort. While the OOT is based on a discrete dynamic model, its self-adaptation is accelerated by a database of physically simulated characteristic buildings, which allows parameter setting at the beginning by a similarity measurement. The OOT artefact provides a base for empirically testing advantages of different SHHC design alternatives.