Undermind is an AI co-researcher for scientific literature search, built for academic researchers, industry R&D scientists and enterprise research teams. A user describes a research question, Undermind asks follow-up questions, and its deep search then reads and evaluates hundreds of papers, follows citation trails and key authors, and adapts its strategy until new searches stop finding relevant papers. The vendor states that a typical search finishes in about 2 minutes, and that a search which has not yet found everything says so in its summary and can be extended. Results come back as ranked paper lists and reports with inline citations to the source papers, alongside a chat that can brainstorm research directions, build custom tables and work through full texts. Undermind searches the Semantic Scholar database of more than 200 million articles, which includes PubMed and arXiv, and notifies users when new papers are published in their areas of interest. It was started by Joshua Ramette and Tom Hartke, quantum physics PhDs from MIT, and is a Y Combinator Summer 2024 company.
Undermind publishes an MCP server that lets AI assistants and coding agents run deep searches, search papers, read full-text PDFs and write cited reports in an Undermind workspace. The vendor documents setup for ChatGPT, where Undermind is a published app, Claude, through its connector directory, Claude Code, Codex, Cursor, VS Code, GitHub Copilot, Microsoft 365 Copilot and Teams through Copilot Studio, OpenCode, Grok, Antigravity and Perplexity, and its tool list covers search, reading, writing, saved searches and workspace organization. For organizations, Undermind offers programmatic API access, batch processing, custom integrations and sign-in through an existing enterprise identity provider. Its GSK case study states that more than 1,000 GSK scientists use Undermind in their daily work and that GSK's internal AI Scientist platform connects to it through a custom API.
Undermind publishes four plans in US dollars, with separate Academic and Industry prices and a monthly or annual toggle. Free costs $0 and includes chat, deep searches, reports, shared workspaces and connected agents at standard rate limits. Pro costs $16 per month billed annually or $20 monthly on Academic pricing, and $60 per month billed annually or $75 monthly on Industry pricing. It adds the latest AI models, full-text analysis, 10x higher usage limits and unlimited workspaces, files and paper libraries. Team is priced per person at $15 per month billed annually or $18.70 monthly on Academic pricing, and $60 or $75 on Industry pricing, shown at five seats, and adds team management, priority support and centralized billing. Enterprise is priced on request and adds increased compute, sitewide organizational login, onboarding seminars, an admin dashboard and custom terms, SLA and security review.
The Free plan searches abstracts and metadata. Full-text search of open-access and preprint articles, the latest models and the higher usage limits start on Pro, and paywalled PDFs can be read only after a user uploads them to the workspace. Paid plans renew automatically, and the refund policy makes charges non-refundable once a billing period has begun, except where the law requires otherwise.
Undermind suits researchers who need exhaustive coverage of a complex or niche question in the scientific literature, and teams that want that search available inside their own AI agents. For formal screening in a systematic review, Rayyan (from $4.99/mo billed annually) is built for that workflow, and Elicit (from $49/mo billed annually) adds data extraction across papers. For a quick read of what studies conclude on a question, Consensus (from $12/mo billed annually) summarizes findings across papers. For sources beyond the scholarly literature, ChatGPT Deep Research ($20/mo) and Gemini Deep Research search the web.
Current state Q4 2026: Undermind's MCP page publishes a benchmark of its deep search against frontier models running as search agents on 23 research goals. In the vendor's own test, Undermind found 85% of the 20 most relevant papers within ten minutes, against 43% to 50% for the agents tested. Its enterprise page and security overview state that a company's data under an enterprise agreement is never used to train AI models and that its model providers do not retain or train on it. For individual accounts, the terms of use let Undermind use user content to improve its services and, in de-identified or aggregated form, to develop AI and machine learning systems. The security overview lists encryption at rest and in transit, role-based access control with multi-factor authentication, daily backups and a manual penetration test by Oneleet, and Undermind publishes no SOC 2 or ISO 27001 certification. Its privacy policy, effective July 31, 2026, covers transfers from the EEA, the UK and Switzerland under Standard Contractual Clauses.
