Semantic extraction

Extraction is sortie’s core primitive: a URL plus a schema in, validated JSON out. The page is rendered in a real Chromium browser, distilled into a compact LLM-readable outline, and the model is forced to submit data matching your schema.

From the CLI

sortie extract https://books.toscrape.com \
  --schema '{"type":"object","properties":{"books":{"type":"array","items":{"type":"object","properties":{"title":{"type":"string"},"price":{"type":"number"}},"required":["title","price"]}}},"required":["books"]}' \
  --instruction "the list of books on the page"

• --schema accepts inline JSON or @path/to/schema.json.
• --instruction is an optional natural-language hint (“the product list”, “only sold-out items”).
• Other flags: --out <file>, --headful (visible browser), --provider anthropic|openai, --model <m>.

From the SDK

zod schemas are the source of truth — the output type is inferred:

import { z } from 'zod';
import { extract } from '@garrison-hq/sortie';

const Books = z.object({
  books: z.array(z.object({ title: z.string(), price: z.number() })),
});

const { data, usage } = await extract({
  url: 'https://books.toscrape.com',
  schema: Books,
  instruction: 'the books listed on the page',
});
// data is typed as { books: { title: string; price: number }[] }

Why it doesn’t break when layouts change

There are no CSS or XPath selectors to go stale. Each run re-renders and re-distills the live page, and the model locates content by meaning within that snapshot. Outputs that fail schema validation are fed back to the model for correction (up to 2 retries), so transient misreads self-correct instead of failing the run.

PDFs

extract works on PDF URLs too — the document is downloaded through the browser’s request context (cookies apply, so authenticated PDFs work) and converted to text before extraction. See PDF support for caps and details.

Replayable extractions

Any extraction worth running twice can be saved as a named query and replayed across pages — see Saved queries.