jimmy·research
Home/AI Infrastructure/Tavily
EN↔中文
Company

Tavily

Purpose-built web search & extraction API for LLM agents — the default retrieval layer for RAG and agentic workflows, now part of nebius.

1. Core Product / Service

  • Tavily Search API — query → ranked URLs + snippets + LLM-clean extracted text, optimized for grounding agent responses (not for human SERP display). Supports basic (1 credit) and advanced (2 credits) depths.[1]
  • Tavily Extract API — given URLs, returns cleaned, LLM-ready content (boilerplate stripped, markdown-friendly). Pairs with Search for two-step retrieval pipelines.
  • Native SDKs + integrations — first-class adapters in LangChain, LlamaIndex, CrewAI; over 3M monthly SDK downloads cited at acquisition.[2]
  • Agent-friendly latency — independent benchmark put Tavily at ~998ms p50 vs Perplexity Sonar ~11s, making it usable in interactive agent loops.[3]

2. Target Users & Pain Points

  • AI agent builders needing live web grounding without scraping/parsing themselves.
  • RAG pipeline teams who want clean text input rather than raw SERPs.
  • Enterprise AI products (cited customers include IBM, Cohere, Groq) needing a non-Google/non-Bing dependency after Microsoft killed the cheap Bing Search API in Aug 2025 — a dislocation Tavily explicitly rode.
  • Pain points solved: parser fragility, Google/Bing TOS risk, mismatched SERP→LLM formats, citation/source-trace hygiene.

3. Competitive Landscape

Competitor Positioning Strength Weakness vs Tavily
Exa.ai Neural/embedding-based "Google for AIs" Semantic queries (concept search) Less keyword precision, smaller LangChain footprint
Perplexity Sonar API Returns synthesized answer + citations Plug-and-play "answer" output ~11s latency; dual billing (req + tokens); not raw building blocks
Brave Search API Independent index, privacy-friendly Cheapest, not Google/Bing-derived Less agent-tuned; raw SERP-style output
SerpAPI / Serper.dev Google SERP scraping Cheapest Google parity Not LLM-optimized; TOS gray zone; no extract layer
Firecrawl / WebSearchAPI.ai Crawl + extract focus Deeper extraction Newer, smaller integrations base

Tavily's moat: deep LangChain/LlamaIndex/CrewAI integration + RAG-shaped output + sub-second latency. Pricing ($0.008/req on Research plan) is mid-range — cheaper than Perplexity, more than Brave/Serper, justified by extraction quality.[1][3]

4. Unique Observations

  • Founded late 2024, acquired by nebius Feb 2026 — under 18 months from inception to exit. ~$25M raised → up to $400M deal = ~16x return on capital. One of the fastest agent-infra exits on record.[2]
  • The acquisition signals a thesis shift: AI clouds (Nebius, CoreWeave-class) are buying the retrieval layer to vertically integrate "compute + grounding" for agent tenants, mirroring how hyperscalers bundled DBs in the 2010s.
  • Founding insight per local research deep-research-workflow: Tavily benefited from Microsoft sunsetting cheap Bing Search API (2025-08-11), which ejected a cohort of incumbents and let agent-native APIs capture ~30-35% share of agent-Web search calls by early 2026.
  • Strategic context for tavily-search-integration: Tavily is the de facto retrieval primitive in hermes-agent / openclaw stacks; switching cost is real because LangChain tool wrappers normalized around its response schema.
  • Endgame risk: as inference providers (openrouter, together-ai) bundle search natively, the standalone search API becomes a feature — Nebius acquisition is a hedge against exactly this commoditization.

5. Financials / Funding

  • Total funding pre-acquisition: ~$25M (lead: Insight Partners).[2]
  • ARR at acquisition: ~$5M (per GetLatka data cited in local research, 2026-03).
  • Headcount: ~28 at deal close.
  • Acquisition (Feb 10, 2026): $275M cash upfront + up to $125M earnout (cash or Nebius stock) → max $400M total.[2]
  • Implied multiple: ~55x ARR upfront, ~80x with earnout — premium reflecting strategic value to Nebius's agent-cloud thesis, not standalone SaaS metrics.

6. People & Relationships

  • Founders:
    • Rotem Weiss (CEO) — Columbia University, NYC-based; staying on post-acquisition to lead Tavily product inside Nebius.
    • Assaf Elovic (CTO) — IDF Unit 8200 background; previously led large-scale system architecture.
    • Yuval Rozio (co-founder) — third co-founder confirmed in acquisition press.
  • Investors: Insight Partners (lead).
  • Acquirer: nebius (Amsterdam-listed AI cloud, ex-Yandex spin-out).
  • Customer / partner ecosystem: IBM, Cohere, Groq (cited at acquisition); deep integration with LangChain, LlamaIndex, CrewAI.
  • Competitive set: Exa.ai, Perplexity (Sonar API), Brave Search API, SerpAPI/Serper.dev.

Sources

Related

nebiusopenroutersurf
Last compiled: 2026-05-09