This thesis presents a real-time multi-agent framework targeted for soft real-time applications. Such an application involves missions each with a soft deadline. A mission is decomposed into phases. The work in each phase can be performed differently, depending on the remaining time. A tradeoff between the quality of the result and the run time requirement is permitted. To make use of this tradeoff, our framework provides a planner that dynamically manages the selection of solution to be used in a phase, based on its knowledge of the deadline and runtime system state. Therefore, the planner may select a faster solution with inferior results in order to meet the mission deadline. This model leads to a real-time self-planned multi-agent framework that is presented as a set of application programming interfaces for the application designer. An implementation of this framework on the Real-Time JVM is also presented. Use of the framework is illustrated with an example intelligent security monitoring system. Finally, an evaluation of the overhead of the planner and its effect on the overall deadline miss rate is reported. Preliminary experimental studies reveal that the planner overhead is rather insignificant, and the self-planned approach can potentially improve both the quality of the result and the deadline miss rate.