Large amounts of time are often spent on the design, debugging, testing and analysis of multilevel (ML) algorithms. MLIT is a package of subroutines aimed at reducing the efforts involved in the writing and analysis of ML algorithms. To shorten programming time, MLIT provides a data structure and a set of respectively general and specific subroutines. The efforts of analysing procedures may be eased using an interactive facility and a set of analysis subroutines e.g. of performance or Fourier analysis. Examples of MLIT usage to design new numerical techniques for nonlinear eigenvalue, bifurcation and optimization problems are presented. An approach to develop adaptive and robust algorithms using local processing, boundary treatment, sequential or simultaneous processing, different transfers and performance tests is illustrated in these examples.