I will propose a method of command line completion based on a probabilistic model. This method supplements the existing deterministic ones. The probabilistic models that are used, are developed within the context of imprecise probabilities and are the main focus of my research. The models are variants of the imprecise Dirichlet model. They are used to represent the assessments about all possible completions. Additionally, they allow for learning by observing the commands typed previously. Because I use an imprecise probabilistic model, a partial (instead of a linear) ordering of the possible completion actions will be constructed during decision making.