We present fast (~200 s sampling) $\it ugriz$ photometry of the low -mass AGN NGC 4395 with the Liverpool Telescope, followed by very fast (3 s sampling) us, gs, rs, is, and zs simultaneous monitoring with HiPERCAM on the 10.4m GTC. These observations provide the fastest ever AGN multiband photometry and very precise lag measurements. Unlike in all other AGN, gs lags us by a large amount, consistent with disc reprocessing but not with reprocessing in the broad-line region (BLR). There is very little increase in lag with wavelength at long wavelengths, indicating an outer edge (Rout) to the reprocessor. We have compared truncated disc reprocessing models to the combined HiPERCAM and previous X-ray/UV lags. For the normally accepted mass of 3.6 × 105M⊙, we obtain reasonable agreement with zero spin, Rout ~ 1700Rg and the DONE physically motivated temperature-dependent disc colour-correction factor (f$\rm _{col}$). A smaller mass of 4 × 104M⊙ can only be accommodated if f$\rm _{col}=2.4$, which is probably unrealistically high. Disc self gravity is probably unimportant in this low-mass AGN but an obscuring wind may provide an edge. For the small mass, the dust sublimation radius is similar to Rout so the wind could be dusty. However, for the more likely large mass, the sublimation radius is further out so the optically thick base of a line-driven gaseous wind is more likely. The inner edge of the BLR is close to Rout in both cases. These observations provide the first good evidence for a truncated AGN disc and caution that truncation should be included in reverberation lag modelling.