Given the importance of sub-daily extreme precipitation events for the occurrence of pluvial floods, it is a key component in climate change adaptation to quantify the likelihood of such extreme events under current and future climate conditions. Such assessments are usually limited by a lack of sufficiently dense and sub-daily precipitation observations, (ii) high-resolution convection-permitting regional climate model (CPM) simulations that realistically represent sub-daily precipitation extremes, and (iii) statistical methods that allow us to extrapolate extreme precipitation return levels under limited data availability and non-stationary conditions (i.e., climate change) based on the main governing physical processes. We overcome these constraints through the utilization of kilometer-scale hourly radar precipitation estimates (RADKLIM) and spatially disaggregated observed daily temperature data (HYRAS-DE-TAS), and the implementation of a novel CPM ensemble covering the entirety of Germany, obtained from the NUKLEUS project within the BMBF-funded RegIKlim (Regionale Information zum Klimahandeln) initiative. Additionally, we introduce the Temperature-dependent Non-Asymptotic statistical model for eXtreme return levels (TENAX) model, a new approach that integrates daily temperature as a covariate, aligning with observed Clausius-Clapeyron scaling rates. This innovation results in a groundbreaking dataset of hourly extreme precipitation for Germany, marking the first instance of accounting for non-stationary climate conditions, i.e., in a +2K and +3K warmer world. The new dataset contains kilometer-scale hourly precipitation extremes for the return level of a 100-year event. Due to the inherent biases of radar-based estimates compared to ground observations, the precipitation extremes have been bias-adjusted on return level basis using KOSTRA.