API Reference

Search

Hybrid keyword + semantic search across apps, screens, and flows.

Gummble's search is hybrid — keyword matching for exact phrases and vector-space ANN for semantic similarity. You can target apps, screens, or both.

GET /api/v1/search?q=onboarding&type=screen&limit=10
NameTypeDescription
qstringQuery string (required)
typeapp | screen | flow | allResult type filter
limitnumberMax results (default 10, max 50)
semanticbooleanForce semantic-only mode (default: hybrid)
POST /api/v1/search/visual
Content-Type: multipart/form-data

Upload an image and get back the top-K visually similar screens by embedding cosine similarity.

curl -X POST https://api.gummble.com/api/v1/search/visual \
  -H "Authorization: Bearer $TOKEN" \
  -F "image=@screenshot.png" \
  -F "limit=10"

Uploads are validated by magic-byte sniffing (not just MIME). Only JPEG, PNG, and WebP under 8 MiB are accepted.

Response shape

{
  "data": [
    {
      "type": "screen",
      "score": 0.87,
      "item": { ... full Screen object ... }
    }
  ]
}

score ranges 0-1. The hybrid scorer blends BM25 + cosine; tune via the semantic flag if you want pure semantic ranking.