Exa AI

Exa is a search API built for AI systems rather than people. Instead of matching keywords and returning a page of blue links, it searches semantically over an embeddings index and returns cleaned page content ready to drop into a model's context — which is what a retrieval pipeline actually needs.

We use it where an agent has to find things on the live web. The difference from a traditional search API shows up on descriptive queries — the kind where you can describe the thing you want but not the words on the page — and in the fact that you get usable content back rather than a URL you then have to fetch and strip.

Tavily is the closest competitor and is often cheaper for straightforward agent search. Perplexity's API returns a synthesised answer with citations rather than sources to reason over, which is a different job — useful when you want the conclusion, wrong when you want the evidence. Brave Search and SerpAPI give you conventional keyword results at lower cost, still the right tool when you know the exact terms. Firecrawl solves the adjacent problem of extracting content from pages you've already identified, and pairs with any of these rather than replacing them.