Exa

Exa’s neural index — search, contents, find-similar, and answer

View as Markdown

Fronts the neural-search API of Exa. Its sibling endpoints all live on the one host api.exa.ai and are mounted under /exa:

POST /exa/<endpoint> → https://api.exa.ai/<endpoint>

Only POST is mounted. All controls live in the JSON body, which is relayed without modification.

Endpoints

Exactly these endpoints are admitted; anything else is refused with the gateway’s own 404 and never reaches Exa.

EndpointPurpose
searchNeural / keyword web search
contentsFetch contents for result URLs
findSimilarFind pages similar to a URL
answerGenerate an answer with sources

The async /research API on Exa’s side is intentionally left unexposed.

Credential handling

Exa honors two credential channels — an x-api-key header and, alternatively, a bearer token in Authorization. The gateway supplies the former and deletes the latter, closing off both:

  • x-api-key is injected; a client-supplied value is overwritten.
  • Authorization is removed before the request goes upstream — which simultaneously shuts the second Exa auth channel; no key can be smuggled through Authorization.

Examples

Responses arrive as Exa’s JSON, untouched. Field-level request and response detail lives in Exa’s documentation; the gateway modifies neither.