We know very little about primordial curvature perturbations on scales smaller than about a Mpc. I review how mu-type distortion of the Cosmic Microwave Background spectrum provides the unique opportunity to probe these scales over the unexplored range from 50 to $104 Mpc-1$. This is a very clean probe, in that it relies only on well-understood linear evolution. While mu-distortion by itself can constrain the amount of power on small scales, correlations between mu-distortion and temperature anisotropies can be used to test Gaussianity. In particular the muT cross correlation is proportional to the very squeezed limit of the primordial bispectrum and hence measures $f_NL$ local, while mumu is proportional to the primordial trispectrum and measures $tau_NL$. A simple Fisher matrix forecast shows that already with current technologies an interesting bound $\Delta f_NL less than 103$ could be achieved. Remarkably, mu-distortion has an extremely small cosmic variance, which in principle allows one to improve this bound by six orders of magnitude. Download the slides