exa.ai-websearch-api

Send a POST request to https://api.exa.ai/search with your x-api-key header, a query string, type set to auto, and num_results. Add a contents object with text and max_characters to retrieve full page text alongside results.

The fast type is optimized for real-time apps, autocomplete, and quick lookups with the lowest latency. The auto type balances relevance and speed at roughly one second and works well for most queries, making it the recommended default.

Yes, Exa works as a tool in OpenAI function calling and Anthropic tool use flows. Define an exa_search tool with a query parameter, execute the search when the model requests it, and return results so the LLM can ground its response in current sources.

Yes, the category parameter filters results to dedicated indexes including people, company, news, research paper, and tweet. Categories can be restrictive, so try searching without one first if you get too few results.

Irrelevant results usually come from overly broad queries or restrictive filters. Use type auto, refine the query with singular specific terms, remove date or domain restrictions, and only add a category after confirming baseline results.

Use the maxAgeHours parameter to set the maximum acceptable cache age in hours. Set 0 to always livecrawl, -1 for cache only, or omit it for the default behavior that livecrawls only when no cache exists.