カオスマップに戻る

Overview

Tavily is a web search and content extraction API built specifically for AI agents and RAG workflows. It is not a consumer search UI — it is designed to be embedded inside LLM applications. Tavily ships official integrations with LangChain and LlamaIndex and is widely used as the search layer for RAG and agentic apps. It now operates as an independent brand under the AI cloud company Nebius.

Key Features

1. Agent-Optimized Search

Traditional search APIs return a list of URLs and snippets. Tavily instead fetches, parses, and cleans up the top results and returns the body text directly — agents get usable content, not links to scrape.

2. Five Endpoints

Tavily exposes five purpose-built endpoints.

  • Search: search and extract in one call
  • Extract: clean extraction from known URLs
  • Map: discover a site's URL structure and generate a sitemap
  • Crawl: walk a site and pull body content in bulk
  • Research: run multi-step autonomous research

3. Official LangChain / LlamaIndex Integrations

Tavily is available as a first-class Retriever / Tool in both frameworks, meaning just a few lines of code to drop it into a RAG or agent pipeline.

4. Domain and Depth Controls

Whitelists, blacklists, search depth (basic/advanced), and HTML content controls give agent developers fine-grained control over behavior.

5. Scale and Reliability

Tavily serves 2M+ developers, handling 300M+ requests a month at 99.99% uptime with a 180ms median latency. On security it added ISO 27001 in July 2026 alongside its existing SOC 2 Type II, with Zero Data Retention on by default.

Pricing

Tiers run from the free Researcher plan up to Growth by credit volume, and the unit price falls as you commit — $0.008 per credit on pay-as-you-go versus $0.005 on Growth.

PlanMonthlyCredits Included
Researcher$01,000 credits/month
Project$304,000 credits/month
Bootstrap$10015,000 credits/month
Startup$22038,000 credits/month
Growth$500100,000 credits/month
Pay-as-you-go$0.008 per credit
EnterpriseContact salesSLA, custom

Credit Costs by Endpoint

Each endpoint has a fixed rate: Search costs 1 credit basic / 2 advanced, Extract 1–2 per 5 successful URLs, Map 1–2 per 10 successful pages, and Crawl combines Map plus Extract. Only Research is variable — 4–110 credits per request on the mini model and 15–250 on pro.

Typical Use Cases

  • Web search tool for LLM agents (RAG retrieval layer)
  • Real-time grounding for Claude/GPT-based assistants
  • Research agents and news-monitoring backends
  • Content extraction and site-crawl pipelines from known URLs

Strengths

  • Purpose-built for agents: returns clean body text
  • Official LangChain and LlamaIndex support
  • Transparent credit pricing that is easy to budget
  • ISO 27001 / SOC 2 Type II and Zero Data Retention meet enterprise requirements

Weaknesses

  • No consumer-facing UI
  • Limited Japanese documentation
  • Costs scale sharply at volume — a single Research call can burn up to 250 credits

Official

公式サイト