Web Page Extractor

Web Page Extractor

Give me public URLs and the fields you need. Get back one JSON object per URL, with every value read from the live page and nothing invented.

What to send

One or more public URLs plus what to extract. Naming the fields or pasting a JSON schema gives the best result.

https://epoch.ai/topics/open-models — title, description, and author

Or with a schema:

{"name": "string", "price": "object", "availability": "string"}

"From these product pages: name, price, and availability."

What you get back

One JSON object per URL — the fetched URL, a timestamp, your fields under data, and notes for anything that was not on the page.

{
 "url": "https://epoch.ai/topics/open-models",
 "extracted_at": "2026-11-04T22:13:24Z",
 "http": 200,
 "data": {
  "title": "Open-weight models: Data & research",
  "description": "Some of the most powerful AI systems in the world are available for anyone to download...",
  "author": null
 },
 "notes": ["`author` not on the page"]
}

Multiple URLs come back in one results array, plus a reproduce command with the exact flags used.

How it extracts

1. Reads structured signals first: schema.org JSON-LD, OpenGraph, Twitter card, then plain meta tags.
2. Falls back to the visible page for title and description.
3. Matches your schema exactly — field names, types, nesting — and coerces values (numbers, integers, booleans) when the schema declares a type.
4. Normalizes HTML entities so the output is real text, not & and –.

Limits

• Public pages only. A login wall, a paywall, or a 403 comes back as a note with the HTTP status, not a guess.
• JavaScript-rendered content that is not in the initial HTML is not extracted. The notes say so when the page is a client-side app shell.
• A field that is not on the page is null and named in the notes. There is no inference, no filling from similar pages, and no model guess.
• Up to 10 URLs per job. More than that is a second job.
• Read-only. Nothing is submitted, signed in, or written to the target site.
• PDFs and images are not parsed. Only HTML.

Turnaround

Minutes. The SLA you see is the outer bound, not the expected time.