One of the major advances in Human-Computer Interaction has been the integration of predictive language modeling in text input interfaces. The introduction of intelligent text entry systems, proposing for instance word completion and word suggestions, made written communication faster and more efficient. Currently, intelligent text entry systems depend upon a manual selection of the intended word or phrase. Such a necessary motor act inherently impacts communication and possibly represents a system bottleneck and a source of potential errors. This required task might be even more problematic in individuals with motor disorders, where simple motor acts, if ever possible, might require high cognitive and physical effort. In this regard, Brain–Computer Interfaces (BCI) provide alternative non-muscular channels for efficient human-computer and human-machine interactions. I here present a prototype BCI system that exploits advanced predictive writing and online brain decoding to boost written communication. Specifically, a novel system combining an intelligent predictive writing system based on GPT with a Rapid Serial Visual Presentation-based P3-BCI speller is described. The proposed LLM-P3-BCI speller prototype represents an exemplary system possibly enabling rapid decoding of user’s intended characters and words via effective online brain signals classification, with no manual intervention. Prospectively, the integration of large language models with BCI promises to substantially augment communication and control in patients with motor or language disorders as well as in healthy individuals.