This work investigates the application of Beyond Diagonal Intelligent Reflective Surface (BD-IRS) to enhance THz downlink communication systems, operating in a hybrid: reflective and transmissive mode, to simultaneously provide services to indoor and outdoor users. We propose an optimization framework that jointly optimizes the beamforming vectors and phase shifts in the hybrid reflective/transmissive mode, aiming to maximize the system sum rate. To tackle the challenges in solving the joint design problem, we employ the conjugate gradient method and propose an iterative algorithm that successively optimizes the hybrid beamforming vectors and the phase shifts. Through comprehensive numerical simulations, our findings demonstrate a significant improvement in rate when compared to existing benchmark schemes, including time- and frequency-divided approaches, by approximately $30.5\%$ and $70.28\%$ respectively. This underscores the significant influence of IRS elements on system performance relative to that of base station antennas, highlighting their pivotal role in advancing the communication system efficacy.