Proofs in logic can become very big and complex. For problem solving, and to teach logic, it is common the use of proof assistants. A general proof assistant should integrate tools to help users on specifying the logics, the formulas, the sets of rules, and the very strategy to perform (semi) automatic proof search. The Automatic Theorem Provers community is aware of some tools that were designed to fulfill these requirements. However, these tools do not take the (possibly) huge size of a proof. Recent works have pointed out that a good way to achieve shorter proofs is the use of graphs, instead of trees, to represent proofs. This dissertation describes and implements a graph-based virtual machine and a compiler for the production of graph-based theorem provers. Some case studies, standard as well as graph-based theorem prover, are illustrated in order to validate the tool.