Exa is an AI-native web search API purpose-built for LLMs and AI agents. Unlike traditional keyword search engines, Exa uses neural embeddings to understand the semantic meaning behind queries, allowing developers to retrieve highly relevant web content even when target pages do not contain the exact search keywords. The platform offers a unified API for search, content crawling, and deep research, with multiple latency modes including Fast (under 450ms), Auto (around 1s), and Deep (around 10s) to suit different agent workflows. Exa also provides specialized vertical search endpoints for News, Companies, Research papers, People, and Financials, backed by an index of over 70 million companies and 1 billion people profiles. Its Highlights feature extracts the most relevant excerpts from pages, delivering up to 90% token reduction for AI applications, while Structured Outputs and Answer endpoints let developers go from query to grounded, citation-backed responses in a single call. Exa is used in production by Cognition, HubSpot, Monday.com, AWS, Databricks, and Groq to power agent search, retrieval-augmented generation, and enrichment pipelines.