Valyu for AI agents

Most search APIs return links and snippets, and the agent's next move is to fetch the pages. Valyu's web search returns the body text already extracted, and its academic search skips the web entirely and returns the paper. Those are two different tools that happen to share a vendor, and each is the right choice for a specific job an agent does often.

What Valyu does well

Excerpts inline. A Valyu web search returns up to ten ranked results with a full-text excerpt for each. For a research loop that would otherwise search, pick URLs, and scrape them, this collapses three calls into one.

Papers, not pages about papers. Valyu's academic search queries arXiv and PubMed directly and returns the paper text, the authors, and the DOI. When an agent is asked for a study, this is the difference between a citation and a blog post summarizing one.

Intent-ranked academic results. An instructions field of up to 500 characters ranks results by what you are actually after, such as "focus on oncology trials from 2023 onwards", which is more useful than a keyword filter for literature work.

Controls for cost and relevance. A relevance_threshold, date windows, country code, a fast_mode that skips the rewrite and rerank step for lower latency, and a url_only mode when you want addresses without content.

Actions and prices through Vaaya

Priced before the call, billed only on success, under the agent's max_cost_cents.

A real call

{
  "service": "valyu",
  "action": "academic",
  "params": {
    "query": "micropayments mental transaction costs",
    "max_num_results": 5,
    "instructions": "prefer economics papers that cite Szabo or Odlyzko"
  },
  "max_cost_cents": 1
}

Five papers with text, authors, and DOIs for a cent. A ceiling of zero refuses the call at no charge.

When to pick Valyu, and when not to

Pick the academic search whenever the deliverable is a paper, a study, or a clinical trial. Nothing else in the network returns the paper itself.

Pick the web search when the agent will need the body text of the results and would otherwise scrape them; the extra cent over Exa buys the extraction.

Pick Exa instead for broad semantic sweeps where a hundred results matter, for people and company discovery, or when a single cent is the budget and snippets are enough. Pick Perplexity when recency per hit decides the answer. Pick Linkup or Tavily when you want a written answer rather than results.

Gotchas

Web search caps at ten results per call; for wider sweeps use Exa. The PubMed side searches full-text papers by default; set include_abstracts to widen it to the whole abstract corpus. Valyu's own price limit field is not a usable spending fence; the ceiling that matters is Vaaya's max_cost_cents, which is enforced before the call.

Valyu's documentation is at valyu.ai. The deep research recipe that escalates from web search to papers is at /recipes/search-scrape.

Questions

What does the Valyu API do?

Two things. Web search that returns each result with a full-text excerpt already extracted, so there is no follow-up scrape for the body. And academic search that queries arXiv and PubMed directly and returns the paper text, authors, and DOI rather than a link to a page about it.

How much does Valyu cost through Vaaya?

Two cents per web search, up to ten results with excerpts inline. One cent per academic search. Priced before the call runs and billed only on success, under the maximum cost the agent states.

When is Valyu the right choice over Exa?

When the task is a paper or study rather than a web page, Valyu's academic search is the only option in the network that returns the paper text with its DOI. For web search, Valyu's inline excerpts are worth the extra cent when the agent needs body text and would otherwise scrape.