5 credits per page with output_schema: 1 for reading the page and 4 for the extraction, $5.00 → $3.00 per 1,000 pages. A summary adds 4. Pages that fail cost nothing, and an extraction that fails costs only the page.

Which model fills the schema?

Claude Haiku 5.5, by Anthropic. Schemas made only of booleans and enums go to OpenAI's Decisions API, which can't return a value outside the schema. meta.llm reports the provider, model and tokens.

How big can a schema be?

Up to 16 KB of JSON Schema, with a prompt of up to 4,096 characters. Up to 20 URLs per call; the call waits up to 75 seconds by default when an AI step runs.

Can it extract from a whole site?

Map the site with /v1/map, or crawl it with /v1/crawl, then send the pages you need to /v1/extract with your schema, 20 at a time.

Does it read pages behind a login or solve CAPTCHAs?

No. It reads public pages the way a logged-out visitor sees them. Pages that can't be read are reported as failed and not billed, and so are short pages that turn out to be bot checks, login walls or soft 404s.

POST /v1/extract · output_schema

AI Data Extraction API

Describe the data you want as a JSON Schema and point it at up to 20 pages or PDFs. SerpKite reads each one, and Claude fills your schema from the page, so you get typed JSON instead of HTML to parse.

Get a free API key Read the reference

5 credits per page with output_schema (1 for the page + 4 for the JSON); failed pages and extractions are free 2,500 free credits, no card Failed calls are free

Overview

What the AI Data Extraction API does

output_schema on /v1/extract turns the Web Extract API into an LLM extraction API. Each page is fetched (logged out, through proxies, JavaScript rendered when needed), converted to Markdown, and Claude Haiku 5.5 returns the data that matches your JSON Schema in results[].output. An optional prompt says what to pick and how to normalise it ("prices in USD", "null when missing"). Values come only from the page.

A schema whose fields are all booleans or enums (classification: "is this a pricing page?", "which plan tier?") is answered by OpenAI's Decisions API instead, which always returns values that match the schema. summary: true adds a 3–6 sentence summary of each page for +4 credits.

Billing is per page: 1 credit for the page plus 4 for the extraction, 5 in total. A page that fails is listed in failed and costs nothing; an extraction that fails is reported in output_error and only the page is billed.

What you get

request object: urls (deduplicated), format, query.
results[] array: One entry per page read: url (final, after redirects), status_code, title, published_at, metadata, cached, markdown / text / html, and links, images, highlights (text, score, heading) when asked for.
results[].summary / summary_error string | null: With summary: true: the summary, or null and the reason it failed (not charged).
results[].output / output_error any: With output_schema: the JSON value matching your schema, or null and the reason it failed (not charged).
failed[] array: URLs that could not be read: url, error (code, message). Not charged.
meta object: request_id, credits_used, latency_ms, succeeded, failed, and llm (provider, model, tokens) when an AI step ran.

Request

POST /v1/extract

cURL

Python

TypeScript

fetch

PHP

curl https://api.serpkite.com/v1/extract \
  -H "Authorization: Bearer $SERPKITE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"urls":["https://www.example.com/pricing"],"output_schema":{"type":"object","properties":{"plans":{"type":"array","items":{"type":"object","properties":{"name":{"type":"string"},"price_per_month":{"type":"number"},"seats":{"type":"integer"}}}},"free_trial":{"type":"boolean"}}},"prompt":"Monthly prices in USD; null when a plan has no list price."}'

Response (illustrative, placeholder domains)

200 OK · application/json

{
  "request": {
    "endpoint": "extract",
    "urls": [
      "https://www.example.com/pricing"
    ],
    "format": "markdown"
  },
  "results": [
    {
      "url": "https://www.example.com/pricing",
      "status_code": 200,
      "title": "Pricing · Example",
      "cached": false,
      "markdown": "# Pricing\n\n…",
      "output": {
        "plans": [
          {
            "name": "Starter",
            "price_per_month": 19,
            "seats": 1
          },
          {
            "name": "Team",
            "price_per_month": 79,
            "seats": 5
          },
          {
            "name": "Enterprise",
            "price_per_month": null,
            "seats": null
          }
        ],
        "free_trial": true
      }
    }
  ],
  "failed": [],
  "meta": {
    "request_id": "req_01J8ZK4M6Q2V7",
    "credits_used": 5,
    "latency_ms": 5120,
    "succeeded": 1,
    "failed": 0,
    "llm": {
      "provider": "anthropic",
      "model": "claude-haiku-5-5",
      "input_tokens": 2840,
      "output_tokens": 96
    }
  }
}

Every response also carries X-Credits-Used, X-Credits-Remaining, X-Cache and X-Tokens-Estimate headers. Full endpoint reference

Reference

Parameters

The fields /v1/extract accepts in its JSON body.

AI Data Extraction API parameters
Name	Type	Default	Description
`urls` required	string[]	–	1–20 public http(s) URLs of HTML pages or PDFs.
`format`	string	markdown	What each result carries: markdown, text or the raw html. Values: markdown, text, html.
`query`	string	–	What highlights are ranked against (up to 4,096 characters).
`highlights`	integer	0	Passages per page ranked against query, 0–10 (about 600 characters each).
`max_tokens`	integer	–	Trim each page's Markdown or text to about this many tokens (100–100,000).
`include_links`	boolean	false	Webpage: also return the page's outbound links (url, text).
`include_images`	boolean	false	Webpage: also return the page's image URLs.
`max_age`	integer	–	Accept a cached page up to this many seconds old, for 0.5 credit.
`country`	string	–	Fetch the pages as seen from this country (two-letter ISO code).
`timeout`	integer	–	Seconds to wait for the pages, 1–90 (default 50, or 75 with summary or output_schema). A page still loading is listed in failed (upstream_timeout) and not charged.
`summary`	boolean	false	Add a 3–6 sentence AI summary of each page, focused on query when set. +4 credits per page; a failed summary is free.
`output_schema`	object	–	A JSON Schema (at most 16 KB). Each page's data comes back as JSON matching it in output. +4 credits per page; a failed extraction is free.
`prompt`	string	–	Guides the output_schema extraction: what to pick and how to normalise it (up to 4,096 characters). Needs output_schema.

Full reference, error codes and headers are in the /v1/extract reference.

Use cases

What people build with the AI Data Extraction API

Prices and products

Plans, prices and specs from product and pricing pages, as rows for your database.

Lead and company data

Company facts, team size and contact pages turned into typed fields.

Classification

Boolean and enum schemas label pages (docs, pricing, blog, login wall) with schema-valid answers.

Documents to records

PDF filings, papers and manuals become JSON with the fields you name.

Pricing