Tavily is a search API for AI agents. Not a thin wrapper around Google — it’s built from the ground up for LLM workflows.

The problem it solves is simple: language models have a knowledge cutoff, and raw web results are a mess of HTML, ads, and noise. Tavily fetches and filters that before returning it, so the model gets clean excerpts it can actually use. /search is the main endpoint. /extract pulls structured content from a specific URL. /crawl goes deeper. There’s also /research for multi-step tasks, which recently topped some benchmarks.

180ms median latency, 100M+ monthly requests. Works with LangChain, OpenAI, Anthropic, and most agent frameworks out of the box.

Researcher plan is free: 1,000 API credits per month.