AI search (RAG)

The AI search endpoint returns results pre-formatted for RAG pipelines. Instead of raw engine hits, you get clean text chunks with structured metadata, ready to inject directly into an LLM context window.

AI search endpoint

POST/v1/ai/search

Search an index and return RAG-formatted results with clean text, metadata, and source attribution.

Auth: admin or search key
Returns: 200

Request body

{
  "index": "articles",
  "q": "how to deploy next.js",
  "limit": 10,
  "filter": "category = 'tutorials'"
}

Response

{
  "results": [
    {
      "content": "Deploy Next.js to Vercel...",
      "metadata": {
        "publishedAt": 1715000000,
        "author": "Jane"
      },
      "source": {
        "index": "articles",
        "documentId": "doc-42"
      },
      "relevanceScore": 0.98
    }
  ],
  "query": "how to deploy next.js",
  "processingTimeMs": 12,
  "totalResults": 42
}

Parameter	Type	Default	Description
`index` · string	string	Required	Index UID to search
`q` · string	string	""	Search query
`limit` · number	number	20	Maximum number of results to return
`filter` · string	string	–	Optional filter expression

Response format

Each result is a structured object designed for LLM consumption:

Field	What it holds
`content`	Clean text with HTML stripped, whitespace normalized: all string fields from the document concatenated. The order of the fields can differ by engine.
`metadata`	Non-string fields (numbers, booleans, nested objects) preserved as structured data
`source.index`	The index UID this result came from
`source.documentId`	The document's primary key value
`relevanceScore`	The engine's ranking score, on Meilisearch (0 to 1) and Visor indexes only. Typesense and Elasticsearch hits have none, so the field is missing there.

The AI search endpoint counts as a search on your plan. No LLM calls are made; this is pure retrieval with formatting.

Search matches words, not meaning. Agents get the best results with keyword queries.

AI search endpoint#

Response format#

AI search endpoint

Response format