coverredact.com

describe_capabilities

Describe the in-page Cover redaction engine for an agent: what it detects, what it refuses, how a document gets onto the page, what the tools return and what they never return. Read this first. Privacy: the document is processed inside this browser tab (WebAssembly); nothing is uploaded and the document never leaves the page. These tools return counts, ids, categories, state and page-relative geometry only — never any text from the document. Limits: Cover redacts a document completely or not at all. Person names and the structured detections are ONE redaction that runs in two stages (the structured pass, then the ~265 MB in-page name model), not two levels you can get separately: a document whose names could not be checked is REFUSED (error daily_limit | incomplete_redaction) rather than returned partly redacted. Scanned or image-only pages are refused (error needs_ocr) and password-protected PDFs are refused (error encrypted_unsupported): OCR and passwords need the Cover Mac app, https://coverredact.com/. Downloads carry the COVER TRIAL watermark unless this browser is registered with a Cover license (https://coverredact.com/): get_status reports the page-level `watermark` and every document reports its own, so read those rather than assuming either way. How a document gets here: the person using the page drops or opens the PDF; tools cannot accept file bytes or URLs. If get_status lists no document and there is no person to ask - a headless run - call load_sample_document: it puts this page's own built-in sample document on the page so the tools have something to work on. It is the ONLY document a tool can introduce.

Page: /

Input schema

No input properties (empty object).

{
  "type": "object",
  "properties": {},
  "additionalProperties": false
}