Company

Exa

Neural search API built for LLMs — "web search rebuilt for AI," using embeddings to do concept-level search instead of keyword matching.

1. Core Product / Service

Exa (exa.ai, formerly Metaphor) is an AI search infrastructure company that rebuilt web search around embeddings rather than keyword/BM25 ranking [1]. Instead of matching query strings, Exa encodes both query and web content into vectors, enabling semantic retrieval ("find pages about X, not containing the words X").

Capabilities:

  • Search API — neural/semantic search returning LLM-friendly structured results (URL, title, snippet, content).
  • Contents API — extract clean text from URLs for downstream LLM use.
  • Entity/company search — specialized endpoints for firmographic lookups, used in research and enrichment pipelines.

2. Target Users & Pain Points

  • AI agent builders needing live, semantically-relevant web grounding.
  • Research / enrichment pipelines (firmographic, literature, due-diligence lookups).
  • Pain solved: keyword search misses conceptually-related content and returns noisy SERPs; Exa sells "search by meaning."

3. Competitive Landscape

Tool Type Note
Exa Neural/semantic search Concept search, entity endpoints
tavily Tavily Search + extract for agents
firecrawl Firecrawl Crawl + scrape
perplexity Perplexity (Sonar API) Synthesized answers + citations
Brave Search API Independent index Cheaper, less agent-tuned

Exa's differentiation is the semantic retrieval layer — it shines on fuzzy/concept queries where keyword APIs (and even tavily) underperform, at the cost of less keyword precision.

4. Unique Observations

  • "Google for AIs" thesis. Exa's bet is that LLMs need a different search than humans — embeddings over an index tuned for meaning, not ad relevance.
  • Rebrand reflects ambition. Dropping "Metaphor" for "Exa" (10¹⁸) signaled the goal of being the default retrieval substrate at exascale.
  • The search-API layer is consolidating fast — tavily sold to nebius for up to $400M (Feb 2026) — and Exa's independent scale (~$700M valuation) makes it one of the last large standalone search-API plays.

5. Financials / Funding

  • Founded: 2021, by Will Bryk & Jeff Wang (Harvard roommates) [1].
  • Total raised: over $111M; valuation ~$700M as of Sept 2025 [2].
  • Headcount: ~75, San Francisco [1].
  • Backers: Y Combinator and venture investors (full list not independently verified in local notes).

6. People & Relationships

Last compiled: 2026-09-14