Exa AI

Exa AI (formerly Metaphor Systems) is the AI-native search infrastructure platform built for AI agents and large language models needing real-time web search, structured content extraction, deep research with structured outputs, and full-site crawling through a single API. Founded around 2021 by Will Bryk (CEO), Dan McArdle, and Jeffrey Wang in San Francisco, Exa has built a proprietary search engine that indexes over 500 billion URLs using models trained from scratch rather than reselling Google or Bing results. In May 2026, Exa raised $250M in a Series C round led by Andreessen Horowitz at a $2.2 billion valuation. That brought total confirmed funding to $361M across four rounds, with participation from Benchmark, Lightspeed Venture Partners, Y Combinator, and NVIDIA. The platform serves 400,000+ developers and 5,000+ businesses. Named customers include Cursor, Cognition (which powers all web access for the Devin AI developer agent through Exa), HubSpot (2.7 million enrichments completed via an agentic lead engine), OpenRouter, and Monday.com. Exa is API-first with native integrations across major AI agent frameworks: LangChain, LlamaIndex, and OpenAI (direct tool-use support). The platform publishes an official MCP server (endpoint mcp.exa.ai/mcp, listed in the Model Context Protocol registry) enabling any MCP-compatible AI system to access all Exa endpoints directly. Python and JavaScript SDKs are open source under the exa-labs GitHub organization. Enterprise features include Zero Data Retention (configurable automatic purging of all queries and data), Single Sign-On for team authentication, and dedicated support with custom SLAs. Notable integration gap: Exa does not publish native integrations with vector database providers such as Pinecone, Weaviate, or Chroma for hybrid retrieval workflows, and does not offer low-code connectors for non-developer teams. Pricing is usage-based with a permanent free tier. Free provides 20,000 requests per month with no credit card required. Search costs $7 per 1,000 requests with web search, token-efficient page contents via Exa Highlights, and configurable latency from 180ms to 1 second. Deep Search costs $12 to $15 per 1,000 requests for multi-step agent workflows with structured outputs and web-grounded citations. Crawling is billed at $1 per 1,000 pages. Enterprise pricing is custom with dedicated SLAs and custom rate limits. A free student program is available for academic use. Exa is not designed for non-developer use cases or teams needing a consumer research interface. Perplexity AI ($17/mo) provides consumer-facing research with built-in web search. ChatGPT Deep Research ($20/mo) and Gemini Deep Research ($19.99/mo) handle autonomous research through conversational interfaces without developer tooling. Tavily (free tier, usage-based) is the closest direct competitor, with a similar four-capability API stack and a larger confirmed developer base at 2M+. Tavily was acquired by Nebius for $275M in February 2026 and is now a Nebius subsidiary, while Exa remains independent with Andreessen Horowitz backing. SerpAPI and Brave Search API provide traditional search access for non-AI applications. The $250M Series C led by Andreessen Horowitz at a $2.2 billion valuation (May 20, 2026) is the defining recent event, bringing total funding to $361M across four rounds. Recent enterprise hires include the former head of retrieval infrastructure from Meta, the head of search backend from Yandex, and a research team from Google, alongside Marcus Holm (President of LaunchDarkly) as Chief Revenue Officer. Exa Highlights processes up to 25 trillion tokens per week through its specialized content extraction model. The platform holds SOC 2 Type II certification (current audit report and 2026 bridge letter) plus HIPAA compliance, with full documentation on the Trust Center at trust.exa.ai. The Websets product extends the API with structured data collection for lead generation and enrichment workflows.