There are long standing plans to integrate adaptive mesh refinement in OpenLB but the only publicly available implementation of this is back from my undergraduate thesis (2D only).
The main issue for implementing computationally efficient general-purpose mesh refinement is that there is both not a clear “best approach” in LBM (AMR is an active research topic) and that an efficient and widely applicable implementation will require extensive modifications of core aspects of OpenLB’s design. Some colleague of mine will work on this in the coming years but don’t expect anything very soon.
Of course this is only speaking about what we want for mesh refinement as a large scale feature – if your concrete requirements are restricted (e.g. only static refinement within the current cuboid decomposition) an implementation is much easier.