Multiplexed single-cell experiment designs are superior in terms of reduced batch effects, increased cost-effectiveness, throughput and statistical power. However, current computational strategies using genetics to demultiplex single-cell (sc) libraries are limited when applied to single-nuclei (sn) sequencing data (e.g., snATAC-seq and snMultiome). Here, we present CellDemux: a computational framework for genetic demultiplexing within and across data modalities, including single-cell, single-nuclei and paired snMultiome measurements. CellDemux uses a consensus approach, leveraging modality-specific tools to robustly identify non-empty oil droplets and singlets, which are subsequently demultiplexed to donors. Notable, CellDemux demonstrates good performance in demultiplexing snMultiome data and is generalizable to single modalities, i.e. snATAC-seq and sc/snRNA-seq libraries. We benchmark CellDemux on 187 genetically multiplexed libraries from 800 samples (scRNA-seq, snATAC-seq, CITE-seq and snMultiome), confidently identifying and assigning cells to 88% of donors. In paired snMultiome libraries, CellDemux achieves consistent demultiplexing across data modalities. Moreover, analysis of 38 snATAC libraries from 149 samples shows that CellDemux retains more genetically demultiplexed nuclei for downstream analyses compared to existing methods. In summary, CellDemux is a modular and robust framework that deconvolves donors from genetically multiplexed single-cell and single-nuclei RNA/ATAC/Multiome libraries.