This report introduces the eGridGPT concept as the first research effort to virtually support system operators in the power grid control room of the future. It outlines interactions between system operators and eGridGPT. It also explains how eGridGPT becomes trustworthy by describing the training process, validating by digital twin, and recommending a dynamic display based on the operator’s prompt. The short example shows that eGridGPT was able to be trained to produce offline planning case results that were similar to the actual model state estimation.