This looks like an interesting tool that can be very effective on certain model classes with 0-1 variables.
More information at:
I believe this is a local search tool with a simulated annealing framework. From what I understand this method really wants to evaluate many solutions as quickly as possible. The method does not like real hard constraints (i.e. preferably constraints go into the objective with an appropriate cost or penalty when violated). In the words of the above paper: “LocalSolver is not designed for solving hardly-constrained optimization problems”.
I suspect the graph shown in http://localsolver.com/images/content/complexity_curves.png is only representative for problems that are suited for this LS approach (so may be no need yet to throw away our MIP and CP solvers).
This also caught my eye:
Please note that business, trial or academic licenses are not available to any individual or entity having their residence or registered office within USA or Canada. Individuals or entities having their residence or registered office within USA or Canada must strictly refrain from downloading any business, trial or academic license.