Undermind

Undermind is an agentic AI literature-search tool that treats a scientific question the way a methodical PhD student would: it reads through individual papers, follows their citation trails, and iterates over multiple rounds before returning a synthesized report with inline references. Rather than ranking results by keyword relevance like a traditional database, it runs an open-ended search loop that decides what to read next based on what it has already found — which is why a single "deep search" takes minutes rather than seconds.

It's built for researchers, R&D teams, and graduate students who need to map a niche corner of the literature rather than skim the top ten hits. Founded by two MIT quantum-physics PhDs (Joshua Ramette and Tom Hartke) and backed by Y Combinator's S24 batch, the product leans toward technical scientific domains and reports early traction with names like GSK, MIT, Harvard, and Caltech.

What separates it from the broader research-AI pack — Elicit, Consensus, Scite, Perplexity — is the agentic, exhaustive-search framing. Where Consensus surfaces a quick evidence snapshot and Perplexity answers conversationally, Undermind is optimized for recall on a specific question: finding the handful of directly relevant papers buried deep in the citation graph. It reports a measure of how confident it is that it found everything relevant, a useful signal when the cost of missing a key paper is high.