Undermind

Undermind is an AI-powered literature search tool built to solve the recall problem in academic research, finding the relevant papers on a topic that conventional keyword and single-pass semantic search consistently miss. Built by two MIT physics PhDs, it runs an adaptive, multi-step agent that reads and evaluates candidate papers, follows citation trails forward and backward, and adjusts its search strategy based on what it learns, using language models as relevance judges much as a human researcher would while scanning results. The output is a report of the papers it found with an explanation of how each is relevant and how thorough the search was, aimed at researchers doing a serious literature review who need confidence they have not overlooked key work. Studies and reviews of deep-research tools have found Undermind achieves notably higher recall and precision than one-shot search approaches. It has been adopted inside research organizations, including a large pharmaceutical company that made it part of an internal AI research stack used by over a thousand scientists, and it is used by researchers at major universities. Undermind offers a limited number of free searches, with paid subscription and credit-based plans for researchers and teams who run many in-depth searches.