Grounded Search

POST /search/grounded

Get an AI-generated answer to a question, grounded in web sources. Every claim is backed by a citation with a URL.

Costs 1 utility call per request.

Request

{
  "query": "What are common complaints about Salesforce pricing?",
  "sources": ["reddit.com", "news.ycombinator.com"]
}
FieldTypeRequiredDescription
querystringyesThe question to answer
sourcesstring[]noRestrict to specific domains

Response

{
  "answer": "Users frequently cite three main pricing complaints about Salesforce: per-seat costs that escalate rapidly as teams grow, expensive add-ons for features that competitors include by default, and opaque enterprise pricing that requires negotiation...",
  "citations": [
    {
      "title": "Salesforce pricing is out of control",
      "url": "https://www.reddit.com/r/salesforce/...",
      "snippet": "We went from $50/user to $150/user after they bundled..."
    },
    {
      "title": "Ask HN: Affordable Salesforce alternatives?",
      "url": "https://news.ycombinator.com/item?id=...",
      "snippet": "The enterprise tier pricing is intentionally opaque..."
    }
  ]
}
FieldTypeDescription
answerstringAI-generated answer grounded in web sources
citationsarraySources backing the answer
citations[].titlestringSource page title
citations[].urlstringSource URL
citations[].snippetstringRelevant excerpt from the source

When to use grounded search vs. research runs

Grounded search is for quick, factual questions where you need a direct answer with citations. It returns in seconds.

Research runs are for open-ended research where you want deep analysis — themes, quotes, sentiment. They take minutes but return much richer data.