Adobe Acrobat

DIRECTION: Scanned/image-based PDF -> searchable PDF via OCR. Use when the user asks to OCR, make searchable, or process/extract/read a scanned or image-based PDF. If the user wants a scanned PDF summarized or extracted, call pdf_ocr first, then pdf_to_markdown. If the user wants a scanned PDF exported to Word/Excel/PowerPoint, call pdf_ocr first, then pdf_export. Do not refuse preemptively because the PDF may be password-protected, encrypted, or corrupted; call the tool and report its result/error. For already-searchable PDFs where the user only wants reading/extraction, use pdf_to_markdown.

pdf_ocr

Check status of asynchronous PDF operations and retrieve results. Call this immediately whenever a previous tool response contains tracking_id. Required parameters: tracking_id exactly from the prior response, and operation_name exactly equal to the tool name that produced that tracking_id, e.g. pdf_compress, pdf_create, pdf_export, pdf_ocr, markdown_to_pdf, pdf_to_markdown, pdf_page_organize. Do not substitute underlying intents such as pdf_combine, pdf_split, or pdf_rotate_pages for a pdf_page_organize tracking_id. If status is in_progress, relay the user_message/elapsed/remaining time and call again. When final, surface download_url/result values. After a final result, stop unless the user explicitly requested a multi-step chain and this final result is needed for the next step. Do not run unrequested follow-up tools such as pdf_compress, export, split, or combine just because a result is available.

pdf_operation_status

Interactive UI and required LLM entry point for PDF page operations: combine, delete, reorder, rotate, split, redact, and highlight. Always call this first for page operations with user_intent exactly one of: pdf_combine, pdf_delete_pages, pdf_reorder_pages, pdf_rotate_pages, pdf_split, pdf_redact, pdf_highlight. Never call direct page tools yourself. Assets rule: assets must be PDFs/URLs/URNs. If any source attachment is a compatible non-PDF, first convert each non-PDF with pdf_create(file_format=source extension), then pass converted PDFs plus original PDFs here. Combine: use user_intent="pdf_combine" for merge/combine requests after needed conversions. If only one existing file is available and no generated/converted content is being added, do not call this tool; ask for another file/content. Delete/reorder/rotate/split: use the matching user_intent and pass known pages/order/angles/split data in organize_params when available. Do not ask for missing details before opening the UI. Do not call pdf_properties for simple “rotate all pages” or “reorder pages”; call pdf_properties first only when total page count/dimensions/orientation are required to compute the instruction. Highlight: call directly with user_intent="pdf_highlight". Pass provided phrases in redactable_strings when available; otherwise open the UI without pdf_to_markdown. Redact: exact provided strings -> call directly with user_intent="pdf_redact" and redactable_strings. General PII discovery -> pdf_to_markdown first, identify strings, then call this tool with those strings.

pdf_page_organize

Reorganize PDF pages with exact sequence order. UI/backend-confirmed tool only. Do not call this tool directly for user requests. For any request to reorder/rearrange/reorganize pages, call pdf_page_organize first with user_intent="pdf_reorder_pages" and pass any provided sequence in organize_params. If the sequence is missing, still open pdf_page_organize rather than asking for page numbers first.

pdf_reorder_pages

Rotate PDF pages. UI/backend-confirmed tool only. Do not call this tool directly for user requests, even when exact pages and angles are provided. For any request to rotate/fix orientation/turn pages, call pdf_page_organize first with user_intent="pdf_rotate_pages" and pass known pages/angles in organize_params. For simple requests like “rotate all pages,” do not call pdf_properties first; open pdf_page_organize directly. Call pdf_properties first only when page dimensions/orientation are needed, e.g. “rotate landscape pages.”

pdf_rotate_pages

Extract document properties: page count, metadata, PDF version, encryption status, and optional page-level details. Use for ambiguous/inspection requests such as “check this PDF,” “what is this file,” “what are its properties,” page count, metadata, encryption status, or when a page operation truly requires total page count/dimensions/orientation. For general ambiguous inspection, set include_page_level_properties=false. Set true only when page-level sizes/orientations/dimensions are needed for a computed operation. Do not use this before simple page operations such as rotate all pages, reorder pages, delete specified pages, or split by provided ranges; call pdf_page_organize directly.

pdf_properties

Permanently redact specified areas in a PDF. UI/backend-confirmed tool only. Do not call this tool directly when the user provides text to remove. Exact words/phrases to redact -> call pdf_page_organize(user_intent="pdf_redact", redactable_strings=[...]). General PII/sensitive-info discovery -> pdf_to_markdown, identify concrete strings, then pdf_page_organize(user_intent="pdf_redact", redactable_strings=[...]).

pdf_redact

Interactive UI for previewing/viewing/opening/displaying a PDF. Use this as the first routed tool when the user asks to view, preview, open, display, or visually inspect a PDF. Do not call pdf_to_markdown for simple viewing requests. If a PDF asset is already available, open it with this tool immediately; do not ask the user to reattach.

pdf_viewer

DIRECTION: PDF pages -> JPEG or PNG images. Use directly when the user asks to convert/export PDF pages as images or JPG/PNG files, including specific page ranges. Do not use pdf_page_organize first for PDF-to-image conversion. To read text, use pdf_to_markdown; to export to Office, use pdf_export.

pdf_to_image

DIRECTION: PDF/documents -> Markdown text for reading, extraction, summarization, classification, Q&A, and content-derived generation. Use for requests to read, extract, summarize, answer questions about document text, or generate new content derived from a PDF. Do not use for simple view/preview/open/display requests (use pdf_viewer), page operations (use pdf_page_organize), PDF->Office (use pdf_export), PDF->images (use pdf_to_image), or ambiguous file inspection/properties (use pdf_properties). Scanned/image-based rule: if the user says the PDF is scanned/image-based or asks to OCR before extraction, call pdf_ocr first, then this tool. Content-derived insertion rule: for adding/generating a TOC, cover page, title page, summary page, glossary, index, appendix, or similar content to insert/combine with an existing PDF, call this tool FIRST, then markdown_to_pdf, then pdf_page_organize(user_intent="pdf_combine"). Do not start with pdf_properties and do not stop after extraction. General PII redaction: extract text first, identify concrete sensitive strings, then call pdf_page_organize(user_intent="pdf_redact", redactable_strings=[...]). Highlight requests do not require pre-extraction; use pdf_page_organize directly. This tool returns markdown text, not a downloadable file; display the markdown content directly when the final user goal is reading/extraction. include_figures parameter guidance: - Set to TRUE for: slides, presentations, lecture notes; documents with diagrams, charts, formulas, or visual explanations; educational, technical, or scientific PDFs. - Set to FALSE only for: plain text documents (e.g., articles, essays, contracts without diagrams). - Default behavior: if document type is unknown, prefer TRUE. If user explicitly asks for "fast" or "text-only", use FALSE. OUTPUT BEHAVIOR: - This tool returns MARKDOWN TEXT, not a downloadable file. - You MUST render or display the markdown content directly to the user. - DO NOT present any URLs in the response as download links.

pdf_to_markdown

Divide a single PDF into multiple separate output files. UI/backend-confirmed tool only. Do not call this tool directly for user requests. For any request to split/divide/extract page ranges into separate PDFs, call pdf_page_organize first with user_intent="pdf_split" and pass provided split options in organize_params. If a prior chain step produced the PDF, wait for its final result and then call pdf_page_organize; never call pdf_split directly.

pdf_split

Handles low-level storage operations: base64 file ingestion and presigned URL generation. Reserved for internal backend workflows. For user uploads, use document_upload instead.

upload_proxy