ABSTRACT We present the new ARTEMIS emulator suite of high-resolution (baryon mass of 2.23 × 104h−1 M⊙) zoom-in simulations of Milky Way-mass systems. Here, three haloes from the original ARTEMIS sample have been rerun multiple times, systematically varying parameters for the stellar feedback model, the density threshold for star formation, the reionization redshift, and the assumed warm dark matter (WDM) particle mass (assuming a thermal relic). From these simulations, emulators are trained for a wide range of statistics that allow for fast predictions at combinations of parameters not originally sampled, running in ∼1 ms (a factor of ∼1011 faster than the simulations). In this paper, we explore the dependence of the central haloes’ stellar mass on the varied parameters, finding the stellar feedback parameters to be the most important. When constraining the parameters to match the present-day stellar mass halo mass relation inferred from abundance matching we find that there is a strong degeneracy in the stellar feedback parameters, corresponding to a freedom in formation time of the stellar component for a fixed halo assembly history. We additionally explore the dependence of the satellite stellar mass function, where it is found that variations in stellar feedback, the reionization redshift, and the WDM mass all have a significant effect. The presented emulators are a powerful tool which allows for fundamentally new ways of analysing and interpreting cosmological hydrodynamic simulations. Crucially, allowing their free (subgrid) parameters to be varied and marginalized, leading to more robust constraints and predictions.