We present CoGraph, a hybrid data- and knowledge base constructed from proceedings of leading conferences in Cognitive Science, Neuroscience, and AI, allowing for the study of conceptual trends across time; we then illustrate a few select examples of its scope and structure. The corpus spans over four decades of research, encompassing tens of thousands of conference publications including from CCN, COSYNE, the Cognitive Science Society, NeuroIPS, and MathPsych, and many others. The corpus includes conference, publication, author, and affiliation data, as well as titles, keywords, abstracts, and extracted textual entities, which are used to generate temporal co-occurrence networks and temporal knowledge graphs. Conferences constitute ideal candidates of study, as they embody bleeding edge trends in research, serve as public forums for the multidisciplinary exchange of ideas, and do so with relatively low latency. Our work provides a basis for studying the coevolving structures underlying the collective emergence of novel ideas and research trajectories, and for assessing the merit of scientific contributions to an ever-broadening ontological horizon—both to explore the space of conceptual content itself and to analyze its convergence and divergence across time.