For the first time, bases of signals delivered by highly segmented HPGe detectors, for identified hit locations, have been determined in situ, that is in the actual accelerator-target-detection system conditions corresponding to data acquisition during a physics experiment. As a consequence, these bases include all the genuine features and alterations of the signals induced by the experimental setup, e.g. diaphony, electronic response, specificity of individual crystals. The present pulse shape bases were constructed using calibration source data taken at the beginning of the AGATA campaign at GANIL. An experiment performed at GANIL using the AGATA $\gamma$ -ray detector together with the VAMOS spectrometer was used to validate the bases. The performance of the bases when used for pulse-shape analysis has been compared to the performance of the standard bases, composed of pulse shapes generated by a computer simulation used for AGATA. This is done by comparing the Doppler correction capability. The so-called Jacobian method used to generate the in situ bases also produces correlations that can be applied to locate in a direct way (no search algorithm) the location where a $\gamma$ -ray interacted given that only one segment is hit. As about 50% of all pulse-shape analysis is performed on crystals with only one segment hit this will allow for a large reduction in the needed computer power. Different ways to improve the results of this prospective work are discussed.