Exa (formerly Metaphor) is a San Francisco-based applied AI lab founded in 2021 by William Bryk and Jeffrey Wang. It builds a search engine and API designed specifically for AI agents, LLMs, and RAG pipelines. Unlike traditional search engines that rank by keyword density and link graphs, Exa uses neural embeddings to understand query intent, retrieving the most semantically relevant pages from the web. The core product is a search API with multiple tiers suited to different latency and depth requirements. Exa Fast delivers results in under 425ms for real-time copilots and voice agents. Standard search includes full page contents for the first ten results at no extra charge. Deep Search and Deep Search (Reasoning) trade latency (12 to 40 seconds) for higher-quality research-grade retrieval. The API supports category-specific indexes covering 1B+ people profiles, 50M+ companies, and 100M+ research papers. Websets is Exa's no-code agentic product for non-technical users. It accepts plain-English queries like 'US-based longevity startups that raised over $10M' and returns hundreds of verified, enriched results including emails and tags. It is used heavily for lead generation, recruiting, and market research, and Exa claims it retrieves over 20x more correct results than Google on complex structured queries. Exa raised an $85M Series B at a $700M valuation in September 2025, then a $250M Series C valuing the company at $2.2B in 2026. Backers include Benchmark, Lightspeed Venture Partners, Nvidia, and Y Combinator. The company reported $10M in annual revenue in September 2025, up roughly 1,010% year-over-year. SDKs are available for Python (exa_py) and TypeScript/JavaScript (exa-js), and an MCP server (exa-mcp-server) is published on GitHub for direct integration with MCP-compatible AI tools. Key features: - Neural embeddings-based search returning results by semantic meaning, not keywords - Multiple search tiers: Exa Fast (<425ms), Standard, Deep Search, and Deep Search (Reasoning) - Category-specific indexes: 1B+ people, 50M+ companies, 100M+ research papers - Websets: agentic no-code product for structured list extraction from the web - Full page contents included with first 10 results at no extra cost on standard tier - MCP server (exa-mcp-server) for integration with Model Context Protocol toolchains - Python and TypeScript/JavaScript SDKs - 1,000 free requests per month on the free tier
Free: 1,000 requests/month. Standard search: $7/1,000 requests (contents for first 10 results included). Deep Search: $12/1,000. Deep Search (Reasoning): $15/1,000. Pro plan from $40/month. Enterprise: custom pricing. Pricing updated March 2026.