A convex nonsmooth optimization problem is replaced by a sequence of line search problems along recursively updated rays. Convergence of the method is proved and applications to linear inequalities, constraint aggregation and saddle point seeking  indicated.