Undermind is an AI co-researcher for scientific literature. A researcher describes a project, Undermind asks follow-up questions, then searches by reading and evaluating hundreds of papers and following citation trails, including full texts and figures. Reports can be refined, details extracted, and answers traced through in-line citations back to the source paper. It also keeps a living library and notifies users when relevant new papers appear. Sign-up is free, and an enterprise option adds secure cloud, no training on customer data and ownership of IP.
Academic scientists, graduate students, R&D teams and pharmaceutical researchers are the audience, and the page says more than 1,000 scientists at GSK use it. A new PhD student can map an unfamiliar field, a drug discovery team can look for the most promising research directions, and a lab lead can monitor topics of interest without repeating searches.
Typical workflows start with a plain-language description of a research goal, followed by a deep search that averages 2.9 minutes in production. Users then gauge relevance for their own question, sort and filter results, generate custom comparison tables and brainstorm directions with an AI expert that has read the relevant papers. In the company's benchmark on 23 complex queries, its deep search reached 85 percent recall of the 20 most relevant papers at ten minutes, compared with 43 to 50 percent for general AI search agents. Undermind can also be connected to AI assistants through its agent connection feature.
