Tavily

Tavily is a web access layer for AI agents, a single API that gives language models real time search, content extraction, web crawling, and deep research. Unlike a traditional search engine built for humans, Tavily returns clean, structured, LLM ready content, ranked snippets, summaries, and citations rather than raw HTML, which keeps agents grounded in fresh information and reduces hallucinations. Founded in 2024 and headquartered in New York with offices in Tel Aviv and Abu Dhabi, it serves more than a million developers, handles over 100 million requests a month at 99.99 percent uptime, and returns search results with a median latency around 180 milliseconds.

The API exposes a few focused endpoints. Search returns ranked, relevance filtered results, with a basic depth for speed and an advanced depth that scrapes more sources, plus an optional synthesized answer. Extract pulls clean markdown text from up to twenty five URLs per call, stripping ads and navigation. Crawl and Map traverse whole sites, and a Research endpoint runs multi step deep research. Tavily reports leading results on retrieval benchmarks including SimpleQA at 93.3 percent. The company grew up inside the LangChain ecosystem, maintains the official langchain-tavily package, and is the default web search tool in many LangChain and LangGraph templates.

Developers integrate through Python and JavaScript SDKs and an official MCP server, with drop in integrations for OpenAI, Anthropic, Groq, LangChain, and LlamaIndex, and availability on the Databricks MCP Marketplace, JetBrains, and IBM watsonx. In February 2026, Nebius Group, an AI infrastructure company, announced an agreement to acquire Tavily for $275M. The founding team joined Nebius and the product continues under the Tavily brand with the same API, so existing integrations keep working, though it is now part of a larger AI cloud company rather than an independent startup.

Pricing runs on a credit system that reflects operation type and depth: a basic search is one credit, an advanced search two, extract is one credit per five URLs, and map is one credit per ten pages. A free tier provides 1,000 credits a month with no credit card, pay as you go costs about $0.008 per credit, and subscription plans lower the unit cost to roughly $0.005 to $0.0075 per credit, with Project and Enterprise tiers for higher limits. Tavily also offers its service free to students. Because cost scales with call volume and depth, agents that hit search on every loop should price their real mix of calls.