In the last few years, the number of assembled genomes has exploded thanks to the new long read sequencing technologies. However, almost all of these assemblies are haploid – especially when there are more than two haplotypes. Indeed, most assemblers struggle to locate the relatively few inter-haplotype differences among all the sequencing errors. Consequently, highly heterozygous regions are assembled as different sequences (contigs), while slightly heterozygous regions and homozygous regions are assembled as single contigs. Tools are then used to “purge” the assembly of the heterozygously assembled “extra” contigs to obtain the final assembly. This remains unsatisfactory, as all of the information contained in the heterozygosity is lost in the assembly. The first challenge to obtain a phased assembly (i.e. with haplotypes assembled separately) is to detect slightly heterozygous regions among all sequencing errors. This can be done using HiFi reads with their extremely low error rate, but those remain expensive and not as long as the more error-prone ONT reads. Hairsplitter implements a new method that takes as input a contig and the set of high-error-rate sequencing reads and tells if the contig is actually a mix of several haplotypes, and if so outputs the haplotype-specific versions of this contig. Hairsplitter calls SNPs rudimentarly and detects recurring patterns of variants among the reads. Unlike existing techniques, It does not need as an input the number of expected haplotypes, as each recurring pattern of variants spontaneously delineate one group of reads. This makes the method useful to assemble polyploid species (wheat, strawberry…), metagenomic samples, and even slightly different repeated regions. The second challenge is to bridge homozygous regions and identical repeats. Assembled as a single contig, these regions need to be duplicated a posteriori to obtain a phased assembly, while making sure that all heterozygous contigs from the same haplotype end up in the same assembled chromosome. Building on an idea implemented in the hybrid assembler Unicycler, GraphUnzip-LR uses long reads mapped to the assembly graph to untangle the graph, duplicating all the homozygous contigs that need to be duplicated. These two new tools can be combined to obtain from a set of long, error-prone sequencing reads a phased assembly, with no limit on the number of haplotypes.