Skip to main content
Build a literature workflow that separates finding papers from evaluating what they say. Use Search to assemble candidate evidence, Contents for accessible URLs, and DeepResearch when you want a cited synthesis. Already have the papers? Use selected-paper comparison for a complete report-and-CSV recipe. This guide covers discovery and source selection.

Define the scope

Record the question, date range, fields and source types before searching. A methods comparison, a recent-paper scan and a systematic review have different completeness requirements.

Find candidate papers

This example searches academic sources for a methods question. Set VALYU_API_KEY in your environment.
cURL
Use a specific dataset ID when the corpus matters, such as valyu/valyu-arxiv or valyu/valyu-pubmed. Check the catalog for available sources and access. A source preset is a convenient scope, not a promise of exhaustive coverage.

Expand PubMed discovery

PubMed search defaults to papers with available full text. Set include_abstracts=true to expand discovery to the abstract corpus. Results can include an abstract and, where available, relevant full-text chunks.
cURL
An abstract-only result is not full-text access. Increasing response_length changes the output limit; it cannot make unavailable text accessible. Some metadata fields, including DOI and authors, may be absent. Keep missing fields empty rather than inventing a citation.

Compare a selected reading set

For a controlled review, select the papers first. Pass their URLs to DeepResearch and explain whether they are mandatory anchors or the entire intended evidence set. Instructions guide the task; check the output to confirm it actually uses the required material.
cURL
Store the returned ID and poll for completion. Preserve the report’s sources alongside its output.

Keep the evidence inspectable

  • Identify preprints separately from peer-reviewed publications.
  • Deduplicate records using available identifiers, while keeping meaningful versions distinct.
  • Record the query, retrieval time and selected sources for reproducibility.
  • Show whether conclusions rely on abstracts, passages or accessible full text.
  • Check required-paper inclusion and unsupported claims before accepting a synthesis.
For systematic reviews, validate retrieval against known relevant studies and a documented search protocol. Ranked natural-language search alone does not establish exhaustive recall or exact Boolean-query equivalence.

Sciences

Connect literature to trials and other scientific records.

Search reference

Control sources, dates and returned content.