Extract

Pass a URL and a JSON schema, get back typed data. Ilmenite fetches the page, runs an LLM against the markdown representation using your schema as a strict constraint, and returns parsed output.

Endpoint

POST https://api.ilmenite.dev/v1/extract

Request body

• url (string, required) — URL to extract from.
• schema (object, required) — JSON Schema describing the output.
• instructions (string, optional) — extra guidance for the extractor.
• model (string, optional) — fast (default) or accurate.

Example

curl -X POST https://api.ilmenite.dev/v1/extract \
  -H "Authorization: Bearer $ILMENITE_API_KEY" \
  -d '{
    "url": "https://www.apple.com/shop/buy-iphone/iphone-16-pro",
    "schema": {
      "type": "object",
      "properties": {
        "model": { "type": "string" },
        "starting_price_usd": { "type": "number" },
        "storage_options_gb": {
          "type": "array",
          "items": { "type": "number" }
        },
        "colors": { "type": "array", "items": { "type": "string" } }
      },
      "required": ["model", "starting_price_usd"]
    }
  }'

Response

{
  "url": "https://www.apple.com/shop/buy-iphone/iphone-16-pro",
  "data": {
    "model": "iPhone 16 Pro",
    "starting_price_usd": 999,
    "storage_options_gb": [128, 256, 512, 1024],
    "colors": ["Desert Titanium", "Natural Titanium", "White Titanium", "Black Titanium"]
  },
  "confidence": 0.97,
  "latency_ms": 1420
}

Schema support

Standard JSON Schema draft 2020-12 features:

• Primitives: string, number, boolean, null
• Containers: object, array
• Constraints: enum, minimum/maximum, pattern, format
• Composition: anyOf, oneOf

Confidence

Every response includes confidence(0–1). Values under 0.6 usually mean the page didn't actually contain the data you asked for. Treat low-confidence rows as unverified.