Source: Exa search API docs
Exa's reference for its own-index, embeddings-based web search API: search types and latency, content options (highlights, text, summary), filters and freshness control.
Contents
- What: official API documentation (two pages)
- Author / publisher: Exa
- Date: read 2026-11-14 (living pages)
- Original: Search reference, How Exa search works, Pricing
Summary
Exa runs its own web index and returns ranked results plus page content in one request, with modes from a ~300 ms cached lookup to multi-step “deep” research.
Key claims
type:auto(default, ~1 s),lite(~300 ms, $4 per 1,000 requests, cached content only, never fetches live),deep-lite(~4 s, light synthesis),deep(4-15 s, multi-step search with synthesis). Legacyfast,instant,deep-reasoningare superseded.- The query is natural language; docs advise describing subject, source type and period.
numResults1-100 (default 10), no pagination;category(company, news, publication, financial report, people, personal site);includeDomains/excludeDomainsup to 1,200 with path prefixes and wildcards; published-date bounds.- Content:
highlights(query-relevant excerpts, the recommended default),text(full clean body, cap withmaxCharacters),summary(LLM per result, extra cost),outputSchemafor a synthesized answer with grounding and confidence. - Freshness:
maxAgeHours(replaces deprecatedlivecrawl): positive N uses cache younger than N hours,0forces content fetched within 60 seconds,-1cache only. It controls content age, not publication date. - Pricing (exa.ai/pricing, 2026-11), per 1,000 requests with up to 10 results:
lite$4,auto$7,deep-lite$12,deep$12; each extra result $1 per 1,000;/contents$1 per 1,000 pages per content type; AI page summaries $1 per 1,000 pages. - Responses include an estimated
costDollarsbreakdown; aneuralcost line suggests embedding-based retrieval (the docs do not detail the index).