Exa Search API

by Exa

A web search API on Exa’s own index, using embeddings so a natural-language query returns ranked results with page content in one request.

How it works

  • Call: POST /search with a query and a contents option; SDKs exist for Python (exa-py) and JavaScript (exa-js); up to 100 results per request, no pagination
  • Search types: auto (default, about 1 second), lite (lowest latency, about 450 ms, cached content only), deep-lite (about 4 seconds) and deep (multi-step, 4 to 15 seconds); the older fast, instant and deep-reasoning still work, and the docs recommend lite, deep or Exa Agent for new work
  • Content options: highlights (recommended), full text capped by maxCharacters, summaries (extra cost), and outputSchema for a synthesized answer as text or structured JSON with grounding
  • Controls: include and exclude domains, user location, and maxAgeHours for how fresh extracted content must be
  • Other endpoints: Contents for known URLs, Answer, Monitors and the Agent API for long-running research; people and company indexes for sales agents and code search for coding agents

Billing model

Pay-as-you-go with no subscription or minimum. The Free tier gives monthly credits with no card; Enterprise is an annual prepaid plan that includes Zero Data Retention and a higher rate limit, with custom limits, SLAs, custom indexes, volume discounts and invoice billing through sales. Per-endpoint rates are on the pricing page; the pricing page does not mention SOC 2.

Position

Exa’s differentiator is semantic, meaning-based retrieval and an index it owns, in contrast with SERP scrapers such as serper-dot-dev and serpapi. Close competitors are parallel-search-api, tavily-search-api, brave-search-api and linkup.

Sources