We present the development and tuning of a topic-adapted language model for word prediction, which improves keystroke savings over a comparable baseline. We outline our plans to develop and integrate style adaptations, building on our experience in topic modeling to dynamically tune the model to both topically and stylistically relevant texts.