Orthology detection from sequence similarity remains a difficult and computationally expensive problem for gene families with large numbers of gene duplications and losses. REvolutionH-tl implements a new graph-based approach to identify orthogroups, orthology, and paralogy relationships first, and it uses this information in a second step to infer event-labeled gene trees and their reconciliation with an inferred species tree. It avoids using gene trees and species trees upon input and settles for a maximal subtree reconciliation in cases where noise or horizontal gene transfer precludes a global reconciliation. The accuracy of the tool is comparable to competing tools at substantially reduced computational cost.