Air pollution is probably the single largest environment risk to health and urban streets are the localized, relevant hotspots. Numerous studies reviewed the state-of-the-art models, proposed best-practice guidelines and explored, using various software, how different approaches (e.g., Reynolds-averaged Navier–Stokes (RANS), large eddy simulations (LES)) inter-compare. Open source tools are continuously attracting interest but lack of similar, extensive and comprehensive investigations. At the same time, their configuration varies significantly among the related studies leading to non-reproducible results. Therefore, the typical quasi-2D street canyon geometry was selected to employ the well-known open-source software OpenFOAM and to investigate and validate the main parameters affecting LES transient simulation of a pollutant dispersion. In brief, domain height slightly affected street level concentration but source height had a major impact. All sub-grid scale models predicted the velocity profiles adequately, but the k-equation SGS model best-resolved pollutant dispersion. Finally, an easily reproducible LES configuration is proposed that provided a satisfactory compromise between computational demands and accuracy.