Undermind

Undermind inverts the search-engine bargain: instead of instant, shallow results, you write a paragraph describing exactly what you need, and it spends several minutes reading hundreds of candidate papers, scoring each against your question before reporting back. Founded by physicists frustrated with keyword search, it is built for the question every researcher eventually hits: "has anyone actually done this specific thing?"

What we like

The depth is not marketing — on hard, precise queries, Undermind routinely surfaces papers that Semantic Scholar, Google Scholar, and faster AI tools simply miss, particularly work hidden behind different terminology or scattered across disciplines. The transparency is the trust-builder: each result comes with an explanation of why it is relevant, and the coverage estimate — roughly how much of the matching literature it believes it found — is a feature no competitor offers and every systematic reviewer will appreciate. Query understanding is genuinely good; it rewards detailed, paragraph-long questions rather than punishing them. For prior-art checks and novelty claims before committing months to a project, one good deep search can justify the subscription alone.

Where it falls short

The economics of depth cut against casual use: searches take minutes, free allowances are small, and everyday "what's the standard reference for X" questions are better served elsewhere. It is a young product — the surrounding workflow (libraries, extraction, alerts) is thin next to Elicit's, and it is a discovery specialist, not a research workspace. Coverage skews toward fields well-represented in indexed literature, with physics and adjacent sciences the sweet spot.

Verdict

Undermind is the tool you reach for when normal search has failed — and in that moment it is the best available. Keep a subscription if your research regularly involves novelty checks or obscure methods; otherwise use the free searches strategically and lean on Elicit and Semantic Scholar for daily work.