Skip to main content
Extract reads the pages you name and fills your JSON Schema with the values they state.
POST /v1/extract answers once the fields are filled, with an id and the invalidURLs it skipped. Send up to 25 urls. Each page is read up to 50,000 characters, and all pages together up to 150,000. The call can take timeout (30 seconds by default) plus 45 seconds, so set your client timeout above that.
The fields are in data. pricePerMonth is left out because the page does not state it. confidence scores each field from 0 to 1, and warning names pages that were skipped or cut. GET /v1/extract/{id} is free and returns your result for 24 hours, or 1 hour with zeroDataRetention.

Request fields

Send a schema, a prompt, or both. With only a prompt, the model names up to 20 camelCase fields. agent is accepted and ignored. To click through a site, use Agent.

Schema

Pydantic model_json_schema() and zod toJSONSchema() output work as they are, in draft-07 or 2020-12. A schema holds up to 100 top-level fields, 32,000 bytes and 32 levels of nesting.

Errors

A failed call, and one that finds none of your fields, is free. Other codes are on Errors.

Next

  • JSON format: the same fields from one page in a scrape call.
  • Agent: navigate a site to find the answer.
  • Batches: scrape a list of pages in one job.