# Undermind — An AI literature-discovery agent that reads hundreds of full papers and follows citation trails, then reports how much of the literature it actually covered.

> Source: The Agents Index — https://theagentsindex.com/undermind (structured, researched, re-verified)
> Facts last verified: 2026-07-25

Undermind is an AI literature-discovery agent purpose-built for scientific research, not general web search. It was founded in 2024 by CEO Joshua Ramette and CTO Tom Hartke — both MIT-trained physicists, with research backgrounds spanning LIGO gravitational-wave astrophysics, CERN particle physics and quantum computing — and backed by Y Combinator's Summer 2024 batch. Given a research question, it iteratively reads and evaluates hundreds of full-text papers and follows citation trails rather than matching keywords, aiming at the "long-tail" papers a conventional database search misses. It is used by researchers at MIT, Harvard, Caltech, Princeton, Berkeley and Cambridge, and by industry teams including 1,000+ scientists at GSK.

| Fact | Value |
| --- | --- |
| Website | https://www.undermind.ai |
| Pricing | $0 |
| API | No |
| Best for | Individual researchers, R&D teams and institutional scientists doing exploratory literature discovery or a novelty check who want citation-trail-following search beyond keyword matching. |
| Not for | Buyers who need a documented, reproducible search strategy for a formal systematic review or regulatory submission (see Elicit's Systematic Review workflow), point-of-care speed, or programmatic/API access — Undermind offers none of the three. |

## Pricing

| Tier | Price |
| --- | --- |
| Free | $0 |
| Pro | $16 / mo (billed annually) |
| Team | $15 / person/mo (billed annually) |
| Enterprise | Custom |

## Verdict

Undermind is the discovery-focused entry in this category: instead of screening papers you already found against inclusion criteria (Elicit's job) or answering an open-web question (Perplexity, Manus, GC AI's different domains), it tries to find the papers a conventional keyword search misses, by reading full texts and following citation trails, and it tells you roughly how complete that search was via its "discovery curve" model. An independent peer-reviewed review backs the core pitch — it genuinely surfaces papers others miss — but the same review is equally direct about the trade-offs: no reproducible search strategy (its self-described "Achilles heel"), an 8-10 minute turnaround too slow for point-of-care use, and no MeSH integration. It also has no API today, unlike Elicit. Pick Undermind when the job is exploratory discovery or a novelty check on a niche or cross-disciplinary question; pick Elicit instead when you need a documented, repeatable screening process for a formal systematic review.

## How it works

1. Describe a research question in plain language, optionally refining it through follow-up dialogue.
2. Undermind reads and evaluates hundreds of candidate papers' full texts and follows citation trails rather than matching keywords.
3. Results are scored for relevance and delivered as an organized report (8-10 minutes typical) with a citation network and a "discovery curve" completeness estimate.
4. Save the search into a project or paper library, and optionally set a recurring alert for new matching papers.

## Who it's for

- Individual researchers and grad students doing exploratory literature discovery on a niche or cross-disciplinary topic
- R&D and industry science teams (e.g. biotech, pharma) who need to catch long-tail prior art a keyword search would miss
- Institutional research teams that already have access to Elicit or PubMed and want a second, citation-trail-based search pass

## Strengths and weaknesses

- ✓ An independent peer-reviewed product review (Journal of the Canadian Health Libraries Association, August 2025) calls it "a worthy, niche competitor" that genuinely helps researchers find highly relevant papers via full-text, citation-trail search rather than keyword matching.
- ✓ Its "discovery curve" statistical model estimates what percentage of the relevant literature has actually been found — a completeness signal the same independent review highlights as a genuine strength most search tools don't offer.
- ✓ Real institutional adoption, not just a demo: used by researchers at MIT, Harvard, Caltech, Princeton, Berkeley and Cambridge, and by 1,000+ scientists at GSK.
- ✓ Founded by two MIT-trained physicist-researchers (LIGO/CERN and top-4-ranked-thesis quantum-physics backgrounds) who built it to solve their own literature-review problem; backed by Y Combinator's Summer 2024 batch.
- ✗ The same independent review calls the inability to produce a reproducible, exportable search strategy Undermind's "Achilles heel" — a real gap for a formal systematic review or regulatory submission that requires a documented, repeatable search method.
- ✗ A typical search takes 8-10 minutes (too slow for point-of-care or real-time lookup) and the tool has no Medical Subject Headings (MeSH) integration the way PubMed does, per the same review.
- ✗ No public API and no documented third-party integrations — both Elicit and GPT Researcher in this same category offer one, Undermind does not.
- ✗ Independent reviewers describe a real learning curve and sparse documentation, and note usage limits can feel tight even on the paid Pro tier.
- ⚠ No public API and no documented third-party integrations at the time of writing.
- ⚠ Cannot produce a reproducible, exportable search strategy — a real limitation for a formal systematic review or regulatory/clinical work that requires a documented, repeatable search method.

## Key features

- **Multi-hop citation-trail search** — Reads and evaluates hundreds of full papers and follows citation trails, rather than one-shot keyword matching against abstracts.
- **Discovery-curve completeness estimate** — A statistical model estimates roughly what percentage of the relevant literature has actually been found — a completeness signal most keyword search tools don't offer.
- **Match-scored, cited reports** — Each candidate paper gets a relevance score; findings are delivered as an organized report with a citation network, not a flat list.
- **Recurring alerts** — Notifies you when new papers matching a saved research topic are published.
- **Dialogue-based query refinement** — Iteratively narrows a complex or niche research question through follow-up questions, echoing a librarian reference consultation.

## Use cases

- **Novelty and prior-art check** — A researcher scoping a new project asks Undermind to surface everything already published on a niche or cross-disciplinary question before committing to a research direction.
- **Finding papers a manual search missed** — A scientist who has already done a keyword-based literature search runs the same topic through Undermind's citation-trail search to catch long-tail, cross-disciplinary papers a database search style misses.
- **Ongoing topic monitoring** — A lab sets a recurring alert on a research area so new relevant papers surface automatically instead of a manual re-search each month.

## Sources

- https://www.undermind.ai
- https://www.undermind.ai/pricing/
- https://www.undermind.ai/about
- https://www.ycombinator.com/companies/undermind
- https://pmc.ncbi.nlm.nih.gov/articles/PMC12352444/
- http://musingsaboutlibrarianship.blogspot.com/2024/04/undermindai-different-type-of-ai-agent.html