Undermind

Undermind is an AI research assistant for scientific literature search, built by a team the site describes as two quantum physics PhDs from MIT and backed by Y Combinator. Users describe a research goal, and Undermind searches the literature, evaluates papers, and returns reports they can iterate on. It is available as a web app, through an API and MCP connection for agents, and as custom enterprise deployments.

What It Is

Undermind is a deep literature search and research synthesis tool. A researcher explains what they are working on, and Undermind asks follow-up questions to clarify the need. It then reads and evaluates hundreds of papers and follows citation trails. Users can iterate on the resulting reports, dive into full texts, extract details, and refine their understanding.

How the Workflow Runs

The site describes four steps: describe the research, explore the literature, go deeper together, and keep up with new work. In the reports, statements link back through in-line citations to source papers. Users can sort and filter papers by relevance to their question and generate custom tables. Alerts notify users when relevant new papers are published.

Agents, API, and Enterprise Options

Undermind can be connected to agents such as Claude and ChatGPT through MCP. The enterprise page lists a web app, an API with programmatic queries and batch processing, and a custom AI research platform that combines internal and external data sources. Enterprise materials state that customer data is not used to train models, that customers own their inputs and outputs, and that data is encrypted in transit and at rest.

Benchmark Claims

Undermind states its v1 engine delivered 10x better results than Google Scholar, citing its whitepaper. Its own benchmark on 23 research goals reports 73% mean recall at ten minutes for Undermind deep search, versus 29–39% for the four agentic web-search configurations shown. These are vendor-published results.