documentreader

const (
	// ConvertToolName is the name of the convert_document tool.
	ConvertToolName = "convert_document"

	// ConvertDisplayName is the display name of the convert_document tool.
	ConvertDisplayName = "Convert Document"

	// ConvertSearchHint is a hint for tool search functionality.
	ConvertSearchHint = "OCR a scanned PDF, extract a complex slide deck, or transcribe audio through a document reader"

	// ConvertDescription is the description of the convert_document tool.
	ConvertDescription = "Convert a local file to markdown using the configured document reader: OCR, layout analysis, table structure, and audio transcription.\n\n" +
		"## When to use\n\n" +
		"- A scanned/image-only PDF that read_file reports as having little or no extractable text\n" +
		"- A PPTX slide deck — most slide decks are mostly visual (screenshots, diagrams, charts) with only a title or two of real text, so read_file's native extraction routinely \"succeeds\" while missing almost everything that actually matters. read_file already auto-falls-back to the document reader for a PPTX whose extracted text is sparse relative to its slide count, but if you're unsure or read_file returned something that looks thin, don't hesitate to call this directly rather than trying to work around it with other tools (terminal, ad-hoc scripts, per-slide image extraction, etc.) — none of that gets better OCR than a whole-document reader does in one shot\n" +
		"- Audio files (WAV, MP3) that need transcription\n" +
		"- A DOCX/PPTX/XLSX that read_file's native extraction failed on\n\n" +
		"## When NOT to use\n\n" +
		"- Plain text-native PDFs, or text-heavy DOCX/XLSX — use read_file instead, it extracts them natively without the network round-trip to an external document-reader service\n" +
		"- read_file already tries the document reader automatically as a last resort for unreadable or sparse files, so you usually don't need this tool for a first read. Reach for it when you want deliberate control: forcing a reconversion, or converting a file read_file didn't route to documentreader.\n\n" +
		"## Rules\n\n" +
		"- Requires a document reader to be configured (document_reader_url) or injected by the host.\n" +
		"- Pass the whole file in one call (`path` is the deck/document itself, e.g. \"deck.pptx\") — the document reader converts every slide/page in a single request. Never split a deck into individual slide images and OCR them one at a time: it's slower, it's more tool calls, and it throws away the layout/reading-order/table-structure analysis a whole-document reader does for a whole document at once.\n"
)

const (
	// ReadURLToolName is the name of the read_document_url tool.
	ReadURLToolName = "read_document_url"

	// ReadURLDisplayName is the display name of the read_document_url tool.
	ReadURLDisplayName = "Read Document URL"

	// ReadURLSearchHint is a hint for tool search functionality.
	ReadURLSearchHint = "fetch and convert a remote document or web page to markdown"

	// ReadURLDescription is the description of the read_document_url tool.
	ReadURLDescription = "Fetch a document at a URL and convert it to readable markdown. " +
		"Supports PDF, DOCX, PPTX, XLSX, HTML pages, and arXiv papers. " +
		"Requires a document reader to be configured. " +
		"Optionally saves the extracted markdown to a workspace path."
)

const (
	RenderPageToolName = "render_document_page"

	RenderPageDisplayName = "Render Document Page"

	RenderPageSearchHint = "render one PDF page as an image for visual inspection of diagrams, charts, tables, or scanned content"

	RenderPageDescription = "Render one page of a local document to a PNG image for visual inspection, then attach it to the conversation as an image (in the message that follows this tool's result) so you can actually see it.\n\n" +
		"Use this after reading a document when markdown extraction may miss important visual details: diagrams, charts, screenshots, equations, scanned pages, or visually structured tables.\n\n" +
		"Current local support is PDF pages through the configured native document renderer. The page number is 1-indexed."
)