Abstract: An electronic device can receive user input via voice or text that includes tasks to be performed. A digital personal assistant infrastructure service can control to which registered action provider the task is assigned. Per-task action provider preferences can be stored. If a preferred action provider is not able to complete the task, the task can still be performed by a registered action provider that has appropriate capabilities. Machine learning can determine a user's preferences. Resource conservation and effective user interaction can result.

Abstract: A method for generating a reactive agent definition may include acquiring, by a reactive agent development environment (RADE) tool of a computing device, an extensible markup language (XML) schema template for defining a reactive agent of a digital personal assistant running on the computing device. The RADE tool may receive input identifying at least one domain-intent pair associated with a category of functions performed by the computing device. A multi-turn dialog flow defining a plurality of states associated with the domain-intent pair may be generated using a graphical user interface of the RADE tool. The XML schema template may be updated based on the received input and the multi-turn dialog flow to produce an updated XML schema specific to the domain-intent pair. The reactive agent definition may be generated using the updated XML schema.

Abstract: An electronic device can receive user input via voice or text that includes tasks to be performed. A digital personal assistant infrastructure service can control to which registered action provider the task is assigned. Per-task action provider preferences can be stored. If a preferred action provider is not able to complete the task, the task can still be performed by a registered action provider that has appropriate capabilities. Machine learning can determine a user's preferences. Resource conservation and effective user interaction can result.