Exa

Regular search engines answer to humans and advertisers; Exa answers to models. Its index is built around embeddings, so a query like “blogs that analyze diffusion model latency, published this month” actually works — and responses return the page contents themselves, ready to drop into a context window, rather than ten blue links plus tracking pixels.

That design made Exa a default web-tool inside the agent stack: coding agents, research agents and automation platforms wire it in for live knowledge. Compared with competitors (Tavily, Serper, Brave's API), Exa leans hardest into semantic retrieval and content delivery. The trade is scale economics: heavy usage costs real money per million queries, and niche-language coverage trails Google's crawl. But for anyone building AI that needs to know the web, Exa is infrastructure.