Ongoing global warming is rapidly moving us away from the climate states we are used to and understand from observational time series. This poses the urgent need to extend our “climate memory” by deciphering climate information stored in geologic archives. However, obtaining quantitative estimates for past climate changes is challenging, particularly in terrestrial archives due to common limitations in time coverage, resolution, and chronology. FluidMICS will employ a novel technique to reconstruct past temperatures based on physical properties of relict drip water preserved in cave formations (speleothems). The behavior of such micrometer-scale fluid inclusions during cooling and heating is directly related to the temperature at which the inclusions were once closed off. This physical basis makes the method uniquely robust and distinguishes it from other paleo-thermometers that depend on empirical calibrations. Combined with the distinct advantages of speleothems, which cover long time periods and can be absolutely dated, this approach will lead to unprecedented insights into magnitude, timing, and distribution of past temperature changes, lifting paleoclimate research to a new level. Our pilot data show that the microthermometry method faithfully discloses past temperatures several hundred thousand years ago. In FluidMICS, we will generate a solid understanding of potential non-thermal effects, further increasing precision and accuracy of the reconstructed temperatures, streamlining the analysis, and extending the applicability of the method. These advances will enable us to generate uniquely accurate and precise terrestrial temperature records far back in time that are distributed over large areas of the globe. These new datasets will serve as invaluable resources to better understand the complexities of our climate system under different atmospheric CO2 concentrations and in times of rapid change, and to test climate models used for future projections.