Exa is a web search API for AI agents, made by Exa Labs Inc., which calls itself an applied AI lab. It crawls and indexes more than 500 billion web pages and ranks them with its own embedding models (neural networks that turn text into vectors of numbers, so that pages with similar meaning lie close together), trained on its own GPU cluster.
A program sends a query in plain language and receives a list of results with titles, links and highlights: short excerpts that a separate model picks as the most relevant parts of each page. Searches can run in a fast mode (about half a second) or a deep mode (about ten seconds) that returns structured output following a JSON schema. Filters cover categories such as news, companies, research papers, people and financial data. Other endpoints return full page contents, run asynchronous deep-research tasks with citations (Agent), and repeat a search on a schedule and report new results by webhook (Monitors). Python and JavaScript SDKs and an MCP server are available.
Exa is a hosted service with free monthly credits, then pay per request.