AI search APIs and web retrieval infrastructure

exa

Exa builds web search infrastructure that gives AI applications access to information beyond their training data. Its Search API supports coding assistants retrieving documentation and public code, chat agents answering questions, and business tools finding companies and people. Developers can configure searches for fast retrieval or deeper research with structured outputs and citations. Websets handles complex queries that take minutes to assemble matching results. Exa also powers Firefox’s desktop Smart Window and iOS Quick Answers, supplying live-web information for cited answers.

Co-founders Will Bryk, the CEO, and Jeff Wang brought experience building AI products at Cresta and data infrastructure at Plaid, respectively. Exa’s engineering approach combines its own web crawling and indexing infrastructure with specialized transformer models that represent documents as embeddings for retrieval. It built a custom vector database to support high query volumes and variable computation per search, and fine-tuned embedding models specifically for code search.

In May 2026, the company reported more than 400,000 developers and 5,000 companies using Exa, including Cursor, Cognition and HubSpot. It also announced a $250 million Series C led by a16z at a $2.2 billion valuation. Sacra estimated Exa’s annualized revenue run rate reached $10 million in September 2025.

Explore the recordings

This guide covers the supplied Exa archive: one recording, with several useful paths through it. Descriptions reflect what the speaker presents in the recording; they do not verify today’s product capabilities, performance or market conditions.

Start with the case for agent search

Watch The Search Engine for the Agentic Web — Will Bryk, Exa for the talk’s central argument: AI agents need precise, database-like retrieval and reliable information. Bryk’s forecast that machine-issued searches will exceed human searches in 2026 is a recorded prediction, not an established current fact.

The Search Engine for the Agentic Web — Will Bryk, Exa ↗

Follow the retrieval design

Use the same talk to explore the query-document matching thought experiment and the techniques described for reducing its cost, including precomputed embeddings and keyword combinations. This path is useful for understanding the design rationale presented in the recording.

The Search Engine for the Agentic Web — Will Bryk, Exa ↗

Explore the API and data-access examples

Return to the recording for examples of complex research queries, selective token extraction, structured results and configurable search constraints. Bryk also describes a claimed 200-millisecond endpoint and Exa Connect’s private-data marketplace, illustrated with Similarweb information. Treat these as recorded capability and performance claims; the supplied catalog does not establish their current availability or independently verify them.