Detect Regions

🤖 AI Drafted (Not reviewed)

Finds WHERE the words are, on-device and free. Apple Vision reads the page locally to locate line and word boxes, so it does produce text — that text is a by-product of finding the boxes, not a transcription: it never replaces the page’s transcript, and no model is called. Runs before a transcriber so every box exists up front.

Tool id	detect_regions
Category	vision
Uses a language model	no
Needs a generative model	no
Runs over many items	yes
Item handling	elementwise
Structured output	no
Human-verified	yes

What it reads

Port	Type	Required	What it is
Files (files)	files	yes	Image files
Context (context)	any	no	Previous text/transcription
Metadata (metadata)	json	no	Existing metadata
Documents (documents)	json	no	Document metadata

What it emits

Port	Type	Required	What it is
Text (text)	text	—	Raw text response
Value (value)	any	—	Parsed value
Texts (texts)	array	—	Per-item texts
Values (values)	array	—	Per-item values
Results (results)	json	—	Full results
Records (records)	array	—	Per-document text records [{doc_id, text}, …].
Artifacts (artifacts)	json	—	Artifact IDs
Files (files)	array	—	The input files, untouched — this step annotates.
Documents (documents)	json	—	The input documents, untouched — this step annotates.

Options

Option	Type	Default	What it does
language	string	en	Recognition locale hint for Apple Vision (an OCR hint, not a claim about the document).
provider	string	apple	apple: free on-device Vision OCR (measured boxes). vlm: send the page to the vision model chosen in the Run Workflow menu (e.g. an OpenRouter model) and ask IT for word boxes — for hands Apple cannot read. VLM boxes are claimed, not measured; replies whose box text is absent from their own transcription are rejected whole rather than rendered. One of: apple, vlm.

The prompt it sends

This tool does not send a prompt — it does its work in code.