The paper introduces Strainy, a novel algorithm for assembling and phasing bacterial strain haplotypes from long-read metagenomic sequencing data (Nanopore and PacBio). Strainy significantly outperforms existing methods in completeness and accuracy, as demonstrated through benchmarking against simulated and real datasets. The algorithm utilizes a phased assembly graph approach to resolve strain heterogeneity within metagenomic assemblies. Application to a real activated sludge metagenome revealed distinct strain distributions and mutational patterns, highlighting Strainy's potential for studying microbial community evolution. The results show Strainy effectively reconstructs complete strain haplotypes, enabling detailed analysis of intra-species variation, including structural variations and antibiotic resistance gene mutations.