In the big data era, sensing multi-modal information in memory is highly demanded for the sake of artificial intelligence applications to overcome the limitations of the von Neumann architecture. Different from traditional sensing methodologies, mechanoluminescence (ML) materials, which emit light in response to mechanical force without any external power supply, present intriguing prospects for technological developments. However, most of the ML materials only demonstrate instantaneous luminescence, severely hampering the exploitation of ML in sophisticated applications where non-volatile control is indispensable. Herein, a non-volatile, multilevel mechano-optical memory system is proposed, based on a crafted combination of a self-recoverable ML material, ZnS:Cu, and a photostimulated luminescence (PSL) phosphor Ca