Exa Search
Run a neural semantic web search from a natural-language query and get back the most relevant results.
query and get back the most relevant results, ranked by semantic relevance rather than keywords. Tune retrieval with type (from fast up to deep-reasoning), bound the result count with numResults (1–100, default 10), and narrow the set with category, includeDomains/excludeDomains, or a published-date range.
Set contents to pull page text, highlights, or an AI summary inline with each result, and supply an outputSchema (optionally with a systemPrompt) to have Exa synthesize a structured output. Billed at a flat $0.08 per successful request; upstream 4xx responses are not billed. When you already know the exact pages you want, post_exa-contents fetches their full text directly.Authorizations
Bearer authentication header of the form Bearer <token>, where <token> is your auth token.
Body
The natural-language search query.
"latest evaluation frameworks for AI agents"
Search mode. auto lets Exa choose; the deep modes trade latency for higher-quality retrieval.
auto, fast, instant, deep-lite, deep, deep-reasoning Number of results to return.
1 <= x <= 100Optional hint about the kind of pages to prioritize.
company, research paper, news, personal site, financial report, people Only return results from these domains.
Exclude results from these domains.
Only return results published on or after this ISO 8601 date.
Only return results published on or before this ISO 8601 date.
Content options to return alongside each result.
JSON Schema used to synthesize a structured output from the results.
Instruction that guides how the structured output is generated.
Response
Search completed successfully.