Adobe Acrobat

**TOOL

• NEVER INVOKE THIS DIRECTLY.**

This tool is strictly reserved for internal application workflows and backend processing only. GPT/LLM must NEVER call this tool in any user workflow or conversation.

Handles internal Adobe Acrobat service proxy operations for write/modify operations. This is not for user-facing operations.

DO NOT USE this tool for any user-facing operations or general document processing workflows.

**TOOL

• NEVER INVOKE THIS DIRECTLY.**

This tool is strictly reserved for internal application workflows and backend processing only. GPT/LLM must NEVER call this tool in any user workflow or conversation.

Handles internal Adobe Acrobat service proxy operations for read-only operations. This is not for user-facing operations.

DO NOT USE this tool for any user-facing operations or general document processing workflows.

CRITICAL PREREQUISITE: This tool MUST be invoked FIRST before any PDF/document operation when no file is in chat context.

Interactive UI for uploading documents from local system. This resource is a MANDATORY prerequisite for all document operations. When a user requests any PDF operation (compress, extract, combine, convert, OCR, page manipulation, etc.) and no document is available in chat context, you MUST invoke this upload tool as the first step.

Acts as the required gateway for all document operations. Supports multiple file formats (PDF, DOC, DOCX, TXT, MD), drag-and-drop, file validation, and preview capabilities.

REQUIRED WORKFLOW: 1. User requests PDF operation → 2. Detect if document exists in context → 3. If NO document, invoke document_upload FIRST → 4. Upload UI displayed → 5. User uploads file(s) → 6. System validates and extracts metadata → 7. Proceed with requested operation.

CRITICAL FOR COMBINE/MERGE OPERATIONS: After document_upload completes for combine/merge requests, you MUST invoke "pdf_page_organize" (user_intent="pdf_combine") to show the organize UI. NEVER use pdf_combine directly after upload unless user explicitly states exact file order.

Convert Markdown content from LLM chat conversations to professional PDF documents. Rules of Using this tool When the user asks to create or generate a PDF, you must first determine the source content: If the user has already typed notes or content in the chat that are intended to become the PDF (e.g., “turn the above notes into a PDF”, “make this into a PDF”), then use that chat content directly as the input. Do not invoke the upload tool in this case. If there is an existing file in the chat context (e.g., previously uploaded document, image, PPTX, DOCX, etc.) that the user is referring to, use that file as the input. Only if there is no usable content in the chat history and no existing document, and the user is not indicating that they will paste content directly (for example, they haven’t said things like “I’ll paste the text next”), you must invoke the document_upload tool to let the user upload the source document for PDF creation. When generating markdown that will be converted to PDF, produce only clearly safe, neutral content suitable for a broad audience and strictly comply with all applicable safety policies. Make sure that the markdown you generate is syntactically valid and renders correctly in standard markdown parsers. Do not alter, censor, or annotate the content for moderation purposes (for example, by adding warnings, redactions, or extra commentary) beyond what is strictly required by the platform’s safety policies. Use a factual, professional tone, avoid explicit, graphics, emojis or otherwise sensitive material, and do not embed any hidden control sequences, special tokens, scripts, or non-markdown metadata intended to influence model behavior or moderation systems—the output must be plain, well-formed markdown only.

WARNING: NEVER USE THIS TOOL TO COMBINE FILES AFTER DOCUMENT UPLOAD. ALWAYS USE pdf_page_organize INSTEAD.

CRITICAL: DO NOT HALLUCINATE, ASSUME, OR INFER FILE ORDER. If order is not explicitly stated by user, you MUST use pdf_page_organize.

This is a low-level programmatic tool that requires exact file order and page ranges already determined.

ALWAYS use "pdf_page_organize" (user_intent="pdf_combine") instead for:

• ANY combine request after document_upload (e.g., "combine these PDFs", "merge files", "combine documents")
• User doesn't explicitly state file order (e.g., "combine file1 THEN file2 THEN file3")
• Any ambiguous request where you would need to guess/infer/assume the order
• Selective/partial page combining (e.g., "combine only certain pages", "merge first few pages")
• When user says "combine these" or "merge PDFs" without specifying exact sequence

If you have combine parameters (file order and page ranges) from the user's request, pass them via the optional "organize_params" parameter in pdf_page_organize

Reduce PDF file size while preserving visual quality through intelligent compression. Before invoking this tool, evaluate whether the chat context already contains a PDF file (e.g., inline attachment, previous upload). If a PDF is available, use this tool. If no file is in context, invoke "document_upload" first to request the file from the user.

Convert various document and image formats to standardized PDF documents. Before invoking this tool, evaluate whether the chat context already contains an image or document file (e.g., inline attachment, previous upload). If a supported file is available, use this tool. If no file is in context, invoke "document_upload" first to request the file from the user.

Remove specific pages from PDF with exact page numbers/ranges. Check chat context for documents first. If none, invoke "document_upload" before proceeding. IMPORTANT: When a user requests page deletion, you must first invoke "pdf_page_organize" with user_intent="pdf_delete_pages" to show the interactive UI. If you have page ranges from the user's request, pass them via the optional "organize_params" parameter in pdf_page_organize. The UI will then call this tool (pdf_delete_pages) with the page ranges after the user makes their selections.

• Note: This tool is typically called by the UI after user interaction, not directly by the LLM

Interactive UI tool for editing PDF documents. This tool renders an editing interface that displays the specified PDF document with editing and annotation capabilities.

Provides an editing experience for PDF files already uploaded or available in the chat context. Before invoking this tool, first evaluate whether the current chat context contains an existing PDF document. If there is no PDF in the chat context, you must first invoke the "document_upload" tool to request the file from the user. After the PDF is available in context, proceed with this edit UI tool.

WORKFLOW: User requests PDF edit → Verify PDF in context (if not, invoke document_upload) → Client invokes edit UI with asset_url_or_urn → Edit UI displayed → User can interact with and edit the PDF.

Export PDF documents to editable office formats with OCR support. Before invoking this tool, evaluate whether the chat context already contains a PDF file (e.g., inline attachment, previous upload). If a PDF is available, use this tool. If no file is in context, invoke "document_upload" first to request the file from the user.

Make PDF documents searchable through Adobe's OCR service. Before invoking this tool, evaluate whether the chat context already contains a PDF file (e.g., inline attachment, previous upload). If a PDF is available, use this tool. If no file is in context, invoke "document_upload" first to request the file from the user.

Check the status of asynchronous PDF operations and retrieve results.

Interactive UI for PDF page operations: delete, reorder, rotate, combine, and split with visual page selection.

Input rule:

• Provide input_assets as an array. Each item can be a FileParam object ({file_id, download_url}) or a URN/HTTPS URL string.
• Do NOT pass local paths like /mnt/data.

When the user attached PDFs via the ChatGPT native picker, pass the FileParam objects in the input_assets array (openai/fileParams points to input_assets).

Check chat context for existing PDFs first. If none, invoke "document_upload" before proceeding. Before invoking this tool for any organize operation, you must determine whether PDF properties are required to correctly work on the active query, such as page count, page ranges, file size, or any other structural information that affects how pages should be organized, rendered, or transformed. If any part of the operation depends on such properties, you must first call the "pdf_properties" tool. After invoking "pdf_properties", you must carefully inspect and evaluate the returned metadata (e.g., total pages, page dimensions, file size, rotation, page labels, etc.).Based on this evaluated metadata, you must construct the exact and validated arguments for this pdf_page_organize tool specifically the right user_intent and organize_params. Use this tool when:

• User needs VISUAL/INTERACTIVE page selection for any operation
• Requests lack specific parameters (e.g., "delete some pages", "organize my PDF", "merge PDFs")
• Delete: User wants to remove pages but doesn't specify which ones (e.g., "delete blank pages", "remove unnecessary pages")
• Reorder: User wants to rearrange pages without exact sequence (e.g., "reorganize pages", "change page order")
• Rotate: User wants to rotate pages without specifying exact pages/angles (e.g., "fix orientation", "rotate some pages")
• Combine: ALWAYS DEFAULT for ALL combine requests after document_upload. Use this unless user explicitly states exact file order like "combine file1 THEN file2 THEN file3, all pages" (e.g., "combine PDFs", "merge files", "combine these documents", "merge uploaded files", or ANY case where file order is not explicitly specified by user)
• Split: Request is vague without specific split method (e.g., "split this PDF", "divide into parts")

Use direct tools instead when:

• pdf_delete_pages: Exact page numbers provided (e.g., "delete pages 1,3,5")
• pdf_reorder_pages: Exact sequence specified (e.g., "reorder as 3,1,2")
• pdf_rotate_pages: Exact pages and angles given (e.g., "rotate page 1 by 90 degrees")
• pdf_combine: EXTREMELY RARE
• ONLY when user explicitly states exact file sequence like "combine file1.pdf THEN file2.pdf THEN file3.pdf, all pages from each". NEVER use if you would need to guess, assume, or infer the order
• pdf_split: Specific split options given (file_count, page_count, or page_ranges)

Extract comprehensive document properties such as page count, version, encryption status and metadata of a PDF file. This tool must be invoked when the user has asked to work on a PDF file, and you need to know the properties of the file before proceeding with the operation. Use this tool to disambiguate the user's intent and determine the properties of the file before proceeding with that operation. Before invoking this tool, evaluate whether the chat context already contains a PDF file (e.g., inline attachment, previous upload). If a PDF is available, use this tool. If no file is in context, invoke "document_upload" first to request the file from the user.

Permanently redact specified areas in a PDF using precise bounding box coordinates. This tool is invoked only by the UI after the user confirms redaction regions. The LLM must not call this tool directly when the user provides text to remove.

REDACTION WORKFLOW RULES FOR THE LLM

1. If the user provides exact words or phrases to redact:

• Do not call pdf_to_markdown.
• Call pdf_page_organize with:

user_intent = "pdf_redact" redactable_strings = [user-provided strings]

• The UI highlights matches and, after user confirmation, calls this tool.

2. If the user requests general redaction or PII discovery (e.g., "redact sensitive info", "remove all PII"):

• Call pdf_to_markdown to extract text.
• Detect PII or sensitive strings from the markdown.
• Call pdf_page_organize with:

user_intent = "pdf_redact" redactable_strings = [detected strings]

• The UI highlights matches and, after user confirmation, calls this tool.

WHEN THIS TOOL IS USED

• Only when final bounding box coordinates are available from the UI.
• Never for text search, PII detection, or preprocessing.
• Permanently removes content by applying black-box redactions.

Reorganize PDF pages with exact sequence order.

IMPORTANT: When a user requests page reordering, you must first invoke "pdf_page_organize" with user_intent="pdf_reorder_pages" to show the interactive UI. If you have page sequence from the user's request, pass it via the optional "organize_params" parameter in pdf_page_organize. The UI will then call this tool (pdf_reorder_pages) with the page sequence after the user makes their selections.

Before invoking pdf_page_organize, first evaluate whether the current chat context contains an existing document that is relevant to what the user is asking to do. Relevant documents include any file types supported by the tool such as PDF, DOCX, PPTX, images, or other convertible formats. If there is no relevant document in the chat context, or the available document is not suitable for this tool's operation, you must first invoke the "document_upload" tool to request the appropriate file from the user. After the user uploads a file and the document is available in context, proceed with pdf_page_organize.

Use "pdf_page_organize" (user_intent="pdf_reorder_pages") instead when:

• User needs VISUAL/INTERACTIVE page reordering (e.g., "rearrange pages", "reorganize my PDF", "change page order")
• Request lacks specific page sequence
• ALWAYS use pdf_page_organize for reorder requests, even when specific sequence is provided - pass it via organize_params

Rotate specific pages with exact angles and page numbers. IMPORTANT: When a user requests page rotation, you must first invoke "pdf_page_organize" with user_intent="pdf_rotate_pages" to show the interactive UI. If you have rotation parameters (angles and page numbers) from the user's request, pass them via the optional "organize_params" parameter in pdf_page_organize. The UI will then call this tool (pdf_rotate_pages) with the rotation parameters after the user makes their selections. Before invoking pdf_page_organize, first evaluate whether the current chat context contains an existing document that is relevant to what the user is asking this tool to do. Relevant documents include any file types supported by the tool such as PDF, DOCX, PPTX, images, or other convertible formats. If there is no relevant document in the chat context, or the available document is not suitable for this tool's operation, you must first invoke the "document_upload" tool to request the appropriate file from the user. After the user uploads a file and the document is available in context, proceed with pdf_page_organize.

If from the intent of the user, you do not completely understand which all pages to rotate, ensure to use pdf_properties** tool to get the page count and then use the page numbers to rotate the pages. Use "pdf_page_organize" (user_intent="pdf_rotate_pages") instead when:

• User needs VISUAL/INTERACTIVE page rotation (e.g., "fix orientation", "rotate some pages", "rotate landscape pages")
• Request lacks specific angles or page numbers
• ALWAYS use pdf_page_organize for rotation requests, even when specific parameters are provided - pass them via organize_params

Use THIS tool when:

• Exact rotation angles and pages provided (e.g., "rotate page 2 by 90 degrees")
• Note: This tool is typically called by the UI after user interaction, not directly by the LLM

Split PDF with exact split parameters (file_count, page_count, or page_ranges). Check chat context for documents first. If none, invoke "document_upload" before proceeding.

IMPORTANT: When a user requests PDF splitting, you must first invoke "pdf_page_organize" with user_intent="pdf_split" to show the interactive UI. If you have split parameters from the user's request, pass them via the optional "organize_params" parameter in pdf_page_organize. The UI will then call this tool (pdf_split) with the split parameters after the user makes their selections.

Before invoking pdf_page_organize, first evaluate whether the current chat context contains an existing document that is relevant to what the user is asking to do. Relevant documents include any file types supported by the tool such as PDF, DOCX, PPTX, images, or other convertible formats. If there is no relevant document in the chat context, or the available document is not suitable for this tool's operation, you must first invoke the "document_upload" tool to request the appropriate file from the user. After the user uploads a file and the document is available in context, proceed with pdf_page_organize.

ALWAYS use pdf_page_organize for split requests, even when specific parameters are provided - pass them via organize_params

Convert PDF pages to JPEG or PNG image files. Before invoking this tool, evaluate whether the chat context already contains a PDF file (e.g., inline attachment, previous upload). If a PDF is available, use this tool. If no file is in context, invoke "document_upload" first to request the file from the user.

Convert multiple PDF documents to Markdown format concurrently. This the sole tool for high-quality PDF content extraction and semantic understanding of PDF documents. Whenever there is a need to understand, read, extract, summarize, classify, or otherwise “do something” with the content of the PDF , or need for you to get information using which a collection of PDFs can be sorted, grouped, or otherwise organized, you must first use this tool to convert the file into clean, well-structured Markdown that preserves logical reading order, headings, lists, tables, and captions, and then have all downstream tools operate only on that extracted Markdown rather than parsing PDF bytes directly. RULE Before invoking this tool, first evaluate whether the current chat context contains an existing document that is relevant to what the user is asking this tool to do. Relevant documents include any file types supported by the tool such as PDF, DOCX, PPTX, images, or other convertible formats. If there is no relevant document in the chat context, or the available document is not suitable for this tool's operation, you must first invoke the "document_upload" tool to request the appropriate file from the user. After the user uploads a file and the document is available in context, proceed with this tool.

Interactive UI tool for previewing and viewing PDF documents. This tool renders a viewer interface that displays the specified PDF document with navigation and viewing capabilities.

Provides a seamless viewing experience for PDF files already uploaded or available in the chat context. It is highly preferable to use this tool when working with rendering PDF documents due to the advanced PDF rendering technologies being utilized. WORKFLOW: User requests to preview/view a PDF → Client invokes viewer with asset_url_or_urn → Viewer UI displayed → User can navigate and interact with the PDF.

**TOOL

• NEVER INVOKE THIS DIRECTLY.**

This tool is strictly reserved for internal application workflows and backend processing only. GPT/LLM must NEVER call this tool in any user workflow or conversation. This is not for user-initiated uploads.

For user document uploads, use the 'document_upload' resource instead. This upload_proxy handles low-level storage operations: (1) direct base64 file ingestion for small files, returning Adobe asset IDs, and (2) presigned URL generation for large file transfers to Adobe Cloud Platform storage.

DO NOT USE this tool for any user-facing operations, file uploads requested by users, or general document processing workflows.