POST

/search

Search for visually similar tiles

Submit one or more queries (text, image, or embedding) and retrieve the top matching tiles from the index.

curl

curl -X POST http://localhost:3000/api/search \ -H "Content-Type: application/json" \ -d '

{
  "queries": [
    {
      "text": "solar system"
    }
  ],
  "n_docs": 5
}

queries

Array of query objectsQuery[]

textoptional

Text search querystring

imageoptional

Base64-encoded imagestring

embeddingoptional

Pre-computed embedding vectornumber[]

n_docsoptional

Number of results to return (default: 20)number

nprobeoptional

FAISS nprobe overridenumber

min_tile_heightoptional

Filter out tiles shorter than thisnumber

instructionoptional

Custom embedding instructionstring

results

One result per queryQueryResult[]

hits

Ranked list of matchesHit[]

score

Cosine similaritynumber

vector_id

FAISS vector indexnumber

article_id

Wikipedia article IDnumber

tile_index

Which 8192px tilenumber

chunk_index

Which 1024px chunk within tilenumber

y_offset

Y position in page (px)number

tile_height

Chunk height (px)number

path

Tile file path on serverstring

url

Wikipedia article slugstring