Skip to main content
Valyu is the search and research layer for AI doing knowledge work. When an assistant or agent needs something it shouldn’t guess, such as a figure from a 10-K, the result of a clinical trial or the claim in a patent, Valyu finds it in the source and hands back the passage with a link. Use it to answer one question in seconds with Search, or to hand over a whole research job with DeepResearch, which comes back with a cited report and the spreadsheet, deck or document you asked for.

Three ways to use Valyu

In Claude or ChatGPT

No code. Add Valyu as a connector and your assistant can search filings, papers and trials.

In your coding agent

One prompt sets up Valyu in Claude Code, Codex or Cursor.

In your product

Call the API from Python, TypeScript or any HTTP client. Your first search takes two minutes.

What makes it different

  • The sources knowledge work runs on. SEC filings indexed by section, full-text papers including licensed Wiley journals, clinical trials, patents, market and economic data, alongside the open web. See the sources.
  • Evidence you can check. Every result carries its source link, and DeepResearch citations can include the exact passage they rest on.
  • Finished work, with sources. DeepResearch returns a cited report plus CSV, Excel, PowerPoint, Word or PDF files, up to ten per task in every mode except instant.
  • Built for agents. The same capabilities through the API, SDKs, a CLI and a hosted MCP server, with docs your agent can read as Markdown.
More on why teams choose Valyu.

See the work

Real DeepResearch reports, published as they came back, each with its CSV and the exact request. Browse all example reports.

Choose an API

All three APIs use the same key and the same balance. Choose by the output your application needs, and combine them freely. Use Search when your software decides the queries, the model and what happens next. For example, a company-research feature retrieves filings and business descriptions, keeps the URLs and passes the evidence to your model. Add Contents to read specific pages you already found. Use DeepResearch when the job spans several questions or sources. Submit a brief, keep the task ID and collect the result when it finishes. For example: compare five companies and return a report and a spreadsheet. If either could fit, run both on a handful of your real questions and compare useful output, latency, cost and the review effort each needs. Evaluate retrieval shows how.

What it costs

You pay for what you use from one credit balance. New accounts start with $10 of free credits ($20 with a company email) and no card.
Free to start$10$20 with a company email
Web search1 creditper 10 results
Contents0.2 creditsper extracted URL
DeepResearch25 creditsper fast task
Plans from $29 a month for 5,000 credits add more credits per dollar and add specialist sources. See pricing for every rate and plan.

Build with the reference

New to these terms? See key terms in plain English. Questions about a source, volume pricing or deployment? Contact us.