Aging is a major risk factor for neurodegenerative diseases, yet the underlying epigenetic mechanisms remain unclear. Here, we generated a comprehensive single-nucleus cell atlas of brain aging across multiple brain regions, comprising 132,551 single-cell methylomes and 72,666 joint chromatin conformation-methylome nuclei. Integration with companion transcriptomic and chromatin accessibility data yielded a cross-modality taxonomy of 36 major cell types. We observed that transposable element (TE) methylation alone distinguished age groups, showing cell-type-specific genome-wide demethylation. Chromatin conformation analysis demonstrated age-related increases in topologically associated domain (TAD) boundary strength with enhanced accessibility at CCCTC-binding factor (CTCF) binding sites. Spatial transcriptomics across 895,296 cells revealed regional heterogeneity during aging within identical cell types. Finally, we developed deep-learning models that reliably predict age-related gene expression changes using multi-modal epigenetic features, providing mechanistic insights into gene regulation. Age-related comparisons use a 2-month baseline reflecting the late-adolescent/early-young adult stage. This dataset advances our understanding of brain aging and offers potential translational applications.