AI Search Engine Rebuilt for LLMs — Neural Retrieval for AI Agents
The consensus on Exa: Choose Exa if you want to build the search infrastructure layer that powers how AI understands the web — hard problems, small team, massive surface area.
Founded in 2021 by Will Bryk (CEO) and Dan McArdle (CTO) — both MIT — Exa started as Metaphor before rebranding to reflect its broader ambition: building the search engine that AI actually needs. Unlike traditional search that matches keywords, Exa uses neural retrieval to understand meaning, letting AI agents search the web the way a human researcher would. Backed by Y Combinator and valued at $700M after an $85M Series B, Exa is the search layer powering AI applications at major labs and enterprises. The culture is engineering-first and fast-moving — a ~120-person team tackling the foundational infrastructure problem of how AI accesses and understands all of human knowledge.
Search rebuilt from scratch for AI. Traditional search engines match keywords — Exa uses neural embeddings to understand meaning, so an AI agent can query "startups working on protein folding" and get real results, not SEO spam.
Custom neural retrieval models trained on billions of web pages. The team builds everything from the crawling infrastructure to the embedding models to the API layer. Try Exa →
Flat and cross-functional. ~120 people, co-founders deeply involved in technical decisions. Engineers own entire verticals — from search indexing to ML model training to API design.
Exa is not an application layer company. You are building the search primitive that thousands of AI agents depend on. If you want your work to be a building block of the AI ecosystem, this is the place.
The company went from founding to $700M valuation in ~4 years, rebranding from Metaphor to Exa along the way. Priorities shift as the AI landscape evolves. You need to be comfortable shipping before things are perfect.
Neural search at web scale is genuinely difficult — indexing billions of pages, training retrieval models that understand intent, serving results in milliseconds. This is not a wrapper on top of someone else's API.
At ~120 people, every engineer's decisions ripple across the product. If you want to own a vertical end-to-end rather than optimize a narrow slice, Exa gives you that scope.