How it works
Literature search without a search server
refract.science runs entirely in your browser. Searching, saving papers and screening them all happen on your device, so your research questions stay yours.
Searched on your device
DuckDB WASM, a database compiled for the browser, answers keyword searches and filters. For semantic search, Transformers.js turns your search into a vector with sciseek-nano, a small model fine-tuned for finding scientific papers. A background thread compares it with every paper's vector and returns the closest papers.
Downloaded once
The venues you install arrive as compact files with titles, abstracts and the papers' vectors (about 160 MB for all 19). They are updated when venues publish new papers. The search model is downloaded on your first visit.
Kept in your browser
Everything is stored in your browser's private storage for refract.science, so the next visit opens without downloading again. Uninstalling a venue frees its space.
AI agents on your terms
Agents like Claude or Cursor search through the Model Context Protocol. A small helper on your computer passes their requests to your refract.science tab, which runs them like your own searches; only the results go to the agent. More on agents
Privacy
- No account, no cookies, no analytics, no advertising.
- Your queries, collections, notes and settings never leave your browser.
- You can choose to share your searches anonymously to support research on literature search. Nothing is sent unless you turn it on, and we never keep identifiable information.
Find details in our privacy policy and terms.