Abstract: Mobile agents is an important enabling technology for certain types of real world applications such as e-commerce and workflows. While the potential benefits are appealing, this technology is not being used by many applications. This is largely attributed to the lack of components such as transactions and their integration with de-facto technologies. This work focuses on bridging this gap and devising an extended transactional model for mobile agents. The nature of mobile agents for autonomous processing and mobility introduces numerous challenges with respect to transactions which are originally addressed in our work. First, given flat, unstructured, execution scripts for agents, it is necessary to isolate side-effects that occur between two successive migrations to later, be able to commit or abort them locally according to transaction semantics maintained across multiple hosts. Second, intra-transaction parallelism requires synchronization among multiple autonomous agents to join their processing and to rollback a transaction. Third, recovery from failures at a specific host, may necessitate maintaining a global state of a transaction across all the agent's destinations to determine how to proceed (e.g., repeat activities upon recovery). This paper presents a comprehensive transaction model for mobile agents and its deployment in the context of Java-based mobile agents. The overall complexity of transaction management is handled using a division of labor between an object-oriented programming model, protocols and a concurrency control mechanism. Specifically, the programming model maintains separation of concerns between mobility, transactions and application logic in order to cope with the aforementioned challenges.