Structured Data Extraction From Any Website (Custom Schema)

{
  "type": "object",
  "required": [
    "url",
    "schema"
  ],
  "properties": {
    "url": {
      "type": "string",
      "format": "uri",
      "description": "Starting website URL to crawl and extract from."
    },
    "schema": {
      "type": "object",
      "description": "JSON Schema describing the object to return. Generate it from a Zod or Pydantic model, or hand-write it — the upstream service fills exactly this shape.",
      "propertyNames": {
        "type": "string"
      },
      "additionalProperties": {}
    },
    "maxAgeMs": {
      "type": "integer",
      "default": 86400000,
      "maximum": 2592000000,
      "minimum": 0,
      "description": "Reuse a cached result younger than this many milliseconds. Default 86400000 (1 day), max 2592000000 (30 days). Set 0 to always fetch fresh."
    },
    "maxDepth": {
      "type": "integer",
      "maximum": 9007199254740991,
      "minimum": 0,
      "description": "Maximum link depth from the starting URL (0 = only the starting page). Unlimited when omitted."
    },
    "maxPages": {
      "type": "integer",
      "default": 5,
      "maximum": 50,
      "minimum": 1,
      "description": "Maximum number of pages to analyze. Default 5, hard cap 50. Does NOT change the price — extraction is billed per call."
    },
    "factCheck": {
      "type": "boolean",
      "description": "When true, every returned value must be grounded in text stated on the page and unsupported fields come back null/empty. When false (default), reasonable inferences are allowed while verifiable specifics stay faithful to the source."
    },
    "timeoutMS": {
      "type": "integer",
      "maximum": 300000,
      "minimum": 1000,
      "description": "Upstream timeout in milliseconds (max 300000). The upstream service aborts the call with a 408 when exceeded. Keep it below the endpoint's requestTimeoutMs so the provider answers before the platform's own budget expires."
    },
    "waitForMs": {
      "type": "integer",
      "maximum": 30000,
      "minimum": 0,
      "description": "Extra browser wait in milliseconds after page load before the content is captured (0-30000). Useful for JavaScript-heavy pages."
    },
    "stopAfterMs": {
      "type": "integer",
      "default": 80000,
      "maximum": 110000,
      "minimum": 10000,
      "description": "Soft time budget for the crawl phase in milliseconds (10000-110000, default 80000)."
    },
    "instructions": {
      "type": "string",
      "maxLength": 2000,
      "description": "Extraction guidance: which facts to prioritize, how to interpret ambiguous fields."
    },
    "includeFrames": {
      "type": "boolean",
      "default": false,
      "description": "Include iframe contents in the Markdown handed to the extractor. Default false."
    },
    "followSubdomains": {
      "type": "boolean",
      "default": false,
      "description": "Follow links on subdomains of the starting URL's domain. Default false."
    },
    "settleAnimations": {
      "type": "boolean",
      "default": false,
      "description": "Wait briefly for animations to settle before each page is read. Default false."
    }
  },
  "additionalProperties": false
}

{
  "type": "object",
  "properties": {
    "code": {
      "type": "string",
      "description": "Present instead of status/data/metadata when the request errors -- observed as 'HTTP_400' for an unreachable/invalid domain."
    },
    "data": {
      "type": "object",
      "description": "The extracted record. Its keys and value types are whatever the caller's own `schema` input requested for that call -- there is no fixed shape here, by design (observed on the one success fixture as {companyName, tagline, industry}, all strings, matching that call's own schema exactly). Only ever observed as a plain object, never null or an array at the top level.",
      "additionalProperties": true
    },
    "status": {
      "type": "string",
      "description": "Present on a successful call. Observed as 'ok' on the one success fixture."
    },
    "message": {
      "type": "string",
      "description": "Human-readable error message accompanying `code`."
    },
    "metadata": {
      "type": "object",
      "properties": {
        "numUrls": {
          "type": "number",
          "description": "Total URLs the crawl considered."
        },
        "numFailed": {
          "type": "number",
          "description": "Pages that failed to fetch."
        },
        "numBlocked": {
          "type": "number",
          "description": "Pages blocked from fetching (e.g. robots rules)."
        },
        "numSkipped": {
          "type": "number",
          "description": "Pages found but not fetched (e.g. over the page/depth limit)."
        },
        "numSucceeded": {
          "type": "number",
          "description": "Pages successfully fetched and read."
        },
        "maxCrawlDepth": {
          "type": "number",
          "description": "Deepest link level actually reached during this crawl."
        }
      },
      "description": "Crawl statistics for the pages the extraction visited.",
      "additionalProperties": true
    },
    "cache_metadata": {
      "type": "object",
      "properties": {
        "age_ms": {
          "type": "number",
          "description": "Age of the cached result in milliseconds, 0 on a fresh fetch (miss)."
        },
        "status": {
          "type": "string",
          "description": "Observed value: \"miss\" (freshly fetched). \"hit\" is documented but unconfirmed here."
        }
      },
      "description": "Whether this result was served from cache or freshly fetched.",
      "additionalProperties": true
    }
  },
  "description": "The extraction result, as returned at the result path, on a successful crawl, or an error code/message when the request itself fails (e.g. an unreachable domain).",
  "additionalProperties": true
}