Tavily

Search API that returns LLM-ready answers and ranked sources

AI Search & KnowledgeFree planOverseasβ˜…β˜…β˜…β˜…β˜† 4.0

What is Tavily?

Tavily is a search API built for LLM applications rather than for people. A single call takes a query, runs a web search, extracts and ranks the content, and returns a short answer together with the source list and cleaned page text. That collapses the usual chain of search, scrape and clean into one request, which is the job most agent and retrieval pipelines actually need. Depth settings trade latency for thoroughness, and a research mode runs several searches and composes a longer answer. It is a commercial API with a free monthly allowance and usage-based pricing above it.

Last updated: 2026-09-20. This site only provides an index; for exact features, pricing, and licensing, see the official website.

Key features

  • Single endpoint returning an answer plus ranked sources
  • Extracted and cleaned page content ready for a prompt
  • Depth settings that trade latency against thoroughness
  • Research mode that runs multiple searches for a longer answer
  • Python and JavaScript SDKs with framework integrations
  • Domain include and exclude filters on each request

Pros & cons

Strengths

  • One call replaces a search, scrape and clean pipeline
  • Sources come back with the answer, so output stays checkable
  • The free monthly allowance is enough to prototype an agent

Watch out for

  • Usage-based pricing needs forecasting before a public launch
  • Answer quality reflects what the open web happens to contain
  • Latency rises noticeably when a deeper research mode is used

Best for & use cases

ai agents, rag pipelines, research automation and developer search backends

If you're comparing similar products, check the alternatives below, or browse all tools in the AI Search & Knowledge category.

FAQ

How is this different from calling a search engine directly?

It does the extraction and ranking for you and returns text with the sources, so the code you write is one request instead of a pipeline of search, fetch, parse and clean steps.

Do results include the sources?

Yes, each response carries the URLs it used, which matters because a generated answer without citations cannot be verified and should not be trusted inside an automated workflow.

What happens past the free allowance?

Charges follow request volume, and deeper research modes cost more per call. Estimate with realistic queries rather than toy prompts before committing an application to it.