Skip to content

Extract typed knowledge

Canonical tool: occam_extract_knowledge

Extract typed structured fields from a page (e.g. title, price, author) as facts[], driven by the site's playbook knowledge_schema.

When to use

  • You need specific data points, not prose — and the host has a resolvable schema (check with occam_playbook_resolve first).
  • No schema for the host → use occam_transcode and read the markdown.

Parameters

Parameter Type Default Required Description
url string — yes HTTP(S) URL or search handle (same host you'd pass to resolve)
backend_policy string http_then_browser no http, browser, or http_then_browser
session_profile string? null no Headers profile id

Returns

Success envelope:

  • ok: true, url, playbookId, pageClass
  • facts[] — {name, value, selector} — typed fields as extracted; unsigned, as-is from DOM
  • meta.koId, latencyMs, backend?, confidence — heuristic confidence, not a correctness guarantee
  • receipt — extraction telemetry only (confidence, elapsedMs). Not Receipt v1. No contentHash, no signature, not accepted by occam_verify.

Failure envelope: ok: false, url, failureCode, message, playbookId?, pageClass?, partialFacts[]?, agentHints?, latencyMs.

Failure codes

invalid_arguments, workers_unavailable, playbook_not_found, knowledge_schema_missing, page_class_unmatched, knowledge_schema_empty, plus transcode fetch taxonomy. See failure codes.

Example

Call:

{ "url": "https://shop.example/product/123" }

Trimmed response:

{
  "ok": true,
  "playbookId": "shop.example",
  "pageClass": "product",
  "facts": [
    { "name": "title", "value": "Widget Pro", "selector": "h1.product-title" }
  ],
  "confidence": 0.9,
  "receipt": { "confidence": 0.9, "elapsedMs": 850 }
}

The receipt field is telemetry — do not feed it to occam_verify.