Page Extract API

Give a document and a JSON Schema; get back JSON that fits the schema, with the page each row is on. Jobs run in the background: create one, then get it until it is done. Base URL https://api.extract.pageindex.ai/v1.

Before you start

Page Extract uses your Page Search API key and credit, and reads documents you uploaded with Page Search. Upload a PDF as in the quick start and keep its doc_… id. Send the key as a bearer token.

Create a job

POST /extract starts a job and answers it at once, PENDING. Get it until its status isCOMPLETED, FAILED or CANCELLED; most documents take a minute or less, the longest several minutes.

import time, httpx

API = "https://api.extract.pageindex.ai/v1"
headers = {"Authorization": f"Bearer {api_key}"}

job = httpx.post(f"{API}/extract", headers=headers, json={
    "document": "doc_60a07bcef3094b68806070f2",   # uploaded with Page Search
    "configuration": {"data_schema": schema, "tier": "standard", "cite_sources": True},
}).raise_for_status().json()

while job["status"] in ("PENDING", "RUNNING"):
    time.sleep(5)
    job = httpx.get(f"{API}/extract/{job['id']}", headers=headers).raise_for_status().json()

print(job["status"], job.get("extract_result"), job.get("error_message"))

Get a job

GET /extract/{id} answers the job; once it is COMPLETED, with its result.

{
  "id": "ext_7b56d419dd0a4d7eafa68e4c",
  "status": "COMPLETED",
  "document": "doc_60a07bcef3094b68806070f2",
  "configuration": {"data_schema": {…}, "tier": "standard", "cite_sources": true},
  "extract_result": {"company": "…", "segments": [{"name": "…", "revenue_usd_millions": …}, …]},
  "extract_metadata": {"field_metadata": {"/segments/0": {"page": 34}, "/segments/1": {"page": 34}, …}},
  "usage": {"pages": 93, "credits": 2.79},
  "created_at": "2026-11-08T12:07:53Z",
  "updated_at": "2026-11-08T12:08:12Z"
}

List, cancel and delete

GET /extract lists your jobs, newest first, without their results. POST /extract/{id}/cancel stops a job that has not finished; DELETE /extract/{id} deletes a job (and stops it first).

curl "https://api.extract.pageindex.ai/v1/extract?limit=20" -H "Authorization: Bearer $PAGEINDEX_SEARCH_API_KEY"
# {"jobs": [{"id": "ext_…", "status": "COMPLETED", …}, …], "next": "ext_…"}   pass next as ?after= for more

Pricing

A job is paid when it completes, by its pages: $0.03 a page at standard, $0.05 at pro, from the same credit as Page Search. A cancelled or failed job costs nothing. Creating a job needs the credit it will take.

Writing a schema

• Describe each field: what it means, where it is printed, the form to give ("as printed, with the currency sign", "in millions"). The descriptions are the instructions.
• Put rows in arrays of objects: every printed row becomes an item, in the order printed, however many there are.
• Let values be missing: {"type": ["string", "null"]} for a field a row may not have.
• Up to 1 MB of schema; $ref and $defs are fine.

Errors

Errors answer {"error": {"code": "…", "message": "…"}}.