A free, no-login research paper discovery tool. Search arXiv, Semantic Scholar, and Crossref at once, save what you like, and get a ready-made prompt to have any LLM summarize a paper for you.
Live: https://paperfinder.dev
Most AI-powered research tools (Elicit, Consensus, SciSpace, ...) require an account and cap usage on a free tier before pushing you to a paid plan. Paper Finder never asks for an account and never hits a usage limit, because it doesn't call an LLM API itself - summarization is "bring your own": copy a pre-built prompt and paste it into whichever LLM you already use.
- Multi-source search - queries arXiv, Semantic Scholar, and Crossref in parallel, merges the results, and deduplicates papers indexed by more than one source (matching on DOI, arXiv ID, or normalized title).
- Semantic relevance ranking - a small sentence-embedding model runs entirely in your browser (no server round-trip) and re-ranks results by actual relevance to your query, not just publication year.
- Save without an account - saved papers live in your browser's
localStorage. No signup, no server-side database, nothing to leak. - Recommendations - a "Recommended" tab surfaces new papers similar to the ones you've saved, via Semantic Scholar's Recommendations API.
- Copy a summarize prompt - one click copies a prompt (title, authors, abstract, link, and instructions) to your clipboard, ready to paste into Claude, ChatGPT, Gemini, or anything else.
- Graceful degradation - if one upstream API is slow or down, the other two still return results; a banner only appears for a genuine outage, not routine rate limiting.
Next.js (App Router, TypeScript) on Vercel. No database, no auth provider, no server-side LLM calls. Tailwind v4 for styling, next/font/google for fonts, fast-xml-parser for arXiv's Atom feed, and a client-side embedding model (via transformers.js) for semantic ranking.
/api/search -> fans out to arXiv + Semantic Scholar + Crossref (Promise.allSettled,
per-source timeout) -> normalize each into a shared Paper shape
-> dedup/merge -> sorted results
/api/recommendations -> forwards saved-paper IDs to Semantic Scholar's
Recommendations API -> normalized Paper[]
client -> localStorage-backed saved papers, copy-to-clipboard prompt
builder, and an in-browser embedding model that re-ranks
results by semantic similarity to the query
See src/lib/papers/ for the per-source adapters and merge logic, and src/lib/embeddings/ for the client-side re-ranker.
When you search, results first appear sorted by year - instant, no AI involved. In the background, your browser then loads a small sentence-embedding model (Xenova/all-MiniLM-L6-v2, via transformers.js) and uses it to compare your query against each result's title and abstract, re-ranking the list by actual relevance once it's done. You can toggle between "Relevance" and "Year" sorting at any time.
A few things worth knowing about it:
- It's free - the model runs entirely in your browser (WebAssembly/WebGPU), not on a server. There's no API to call and no cost, so this feature can't be metered, capped, or taken away.
- Nothing leaves your device for this - your query and the search results are already in your browser; the model just runs inference locally, no network round-trip involved.
- It's ~90MB, downloaded once - your browser caches it after the first search, so it's only a one-time cost, not a per-search one.
- It degrades gracefully - if your browser can't run it (or the download is blocked), search still works normally, just without the relevance re-sort.
- It also cleans up your query first - common filler words ("a", "that", "for", ...) are stripped before searching arXiv/Semantic Scholar/Crossref, so a loosely remembered description ("that paper about attention without recurrence for sequence stuff") has a much better chance of surfacing the right result even before the AI ranking kicks in.
git clone https://github.com/emmanuel-adu/paper-finder.git
cd paper-finder
npm install
cp .env.example .env.local # fill in CROSSREF_MAILTO at minimum
npm run devOpen http://localhost:3000.
| Variable | Required | Notes |
|---|---|---|
CROSSREF_MAILTO |
Yes | Your email, sent to Crossref's API to opt into their faster "polite pool." Not a secret, just identifies the requester. |
SEMANTIC_SCHOLAR_API_KEY |
No | Optional. The unauthenticated tier shares a global rate-limit pool and can 429 under load. A free key (request one at https://www.semanticscholar.org/product/api#api-key-form) removes that limit. |
Contributions are welcome - see CONTRIBUTING.md for local dev setup and PR guidelines.