seqArchR enables unsupervised discovery of de novo clusters with characteristic sequence architectures characterized by position-specific motifs or composition of stretches of nucleotides, e.g., CG-richness. seqArchR does not require any specifications w.r.t. the number of clusters, the length of any individual motifs, or the distance between motifs if and when they occur in pairs/groups; it directly detects them from the data. seqArchR uses non-negative matrix factorization (NMF) as its backbone, and employs a chunking-based iterative procedure that enables processing of large sequence collections efficiently. Wrapper functions are provided for visualizing cluster architectures as sequence logos.