<p>Cytotoxic T cells (CD8⁺) play a pivotal role in immunosurveillance by identifying and eliminating tumor cells. However, the onset of CD8⁺ T cell exhaustion, characterized by overexpression of immune checkpoint receptors, impairs their function, allowing tumor cells to evade immunosurveillance. Single-cell metabolic profiles hold the promise in characterizing intrinsic cellular metabolic heterogeneity of exhausted CD8⁺ T cells, even at ultra-early time points. Herein, we developed a multi-modal single-cell platform that integrates nanoparticles-enhanced laser desorption/ionization mass spectrometry and protein number counting platform to elucidate the temporal dynamics of CD8⁺ T cell exhaustion. A comprehensive time-series analysis was conducted, with nearly 3000 single cells performing metabolic profile extraction and checkpoint receptors quantification. Our results demonstrated that the onset of exhaustion was as early as 3&#xa0;h post-stimulation and upon cessation of stimulation, a degree of reversibility was observed in these exhausted cells. Using deep learning algorithms, the discrimination of the different exhausted CD8⁺ T cell subpopulations from the control achieved an area under the curve value of more than 0.904, even to 1.000 with 100% sensitivity and specificity. Our work presents a robust, high-throughput, and scalable system for multi-modal single-cell analysis, offering valuable insights into the dynamics of CD8⁺ T cell exhaustion.</p>