Exa is an applied AI research lab building a search engine designed for AI agents and large language models rather than human end users. The company trains, embeds, and serves its own large-scale web index and retrieval models, returning ranked URLs and dense, query-specific excerpts that let AI applications ground their answers on the open web.

Exa offers a developer API for search, content retrieval, and synthesized answers, plus Websets — a no-code research surface — and an Exa Research blog that publishes the lab's work on neural retrieval. Its differentiator is a link-prediction approach to ranking, training the model to predict the next link rather than the next word, which the team positions as a fundamentally different retrieval substrate from keyword-based engines.