Searching for papers

Three ways to find papers; keyword, natural language, and hybrid.

researchwith.ai's search runs against an index of over 250 million papers, drawn from arXiv, OpenAlex, and other open sources. You can search three ways depending on what you know.

The default mode. Type a few words and we'll return the closest title and abstract matches, ranked by relevance.

Best for:

  • You know the paper's title or roughly the topic
  • You want to filter by an author or term

Example: attention is all you need returns the transformer paper near the top.

Type a full question instead of keywords. We use an LLM to interpret the question, generate a search plan, and walk a vector index to find papers that actually address the question, not just match the words.

Best for:

  • You have a research question but don't know which authors work on it
  • You want to compare approaches across papers
  • You're new to a topic
  • Have a complex request that matches the topic, author, time period and more

Example: Show me AI safety research papers written by Geoffrey Hinton since 2021 returns results based on complex query understanding for topic, author and time filters.

When you want the best of both, hybrid blends keyword and vector relevance. It's the default behind the "Search everything" mode and is what powers the search inside a paper.

Rate limits

Searches count against your daily quota:

  • Free: 20 searches per day
  • Pro: 100 searches per day

Quotas reset at midnight UTC.

Tips

  • Wrap a phrase in quotes to require an exact match
  • If you know paper you want to read, skip search entirely. Just add the url or arXiv id in the search box. It can pick up urls vs keywords automatically to open the reader page.
  • The first 5 results are usually the strongest. If none look right, refine the query rather than scrolling
Previous
Paper notes and line notes
Next
Filters and recent searches