A tool that maps intermediate states in vision-language models to a shared semantic space to study concept accessibility across modalities.