What file types, limits, and processing capabilities are supported in AI Agent Studio?

AI Agent Studio provides several options for processing files, including chat uploads, Document Tool ingestion, and background processing with Document Processor.

You can upload these types of files:

• Documents: PDF, DOCX, PPTX
• Text and structured text: TXT, HTML, MARKDOWN, JSON, XML
• Spreadsheets and tabular data: XLSX, CSV
• Images: JPG, PNG
• Multimedia: WAV, MP3, M4A, AAC, OGG, Opus, WebM, FLAC, MP4
• Archives: ZIP
  Note: ZIP files are supported only when using RAG Document Tool, and multimedia files aren't supported for RAG Document Tool.

This table outlines the recommended usage patterns for chat uploads, Document Tool, and the Document Processor. By understanding these capabilities, limitations, and best practices, you can optimize file processing performance and ensure reliable handling of uploaded content across different processing flows.

Recommended patterns

Goal	Option to use	Best suited for	Key limitations	Can you choose LLM?
Work with a file during a chat conversation	Chat file upload	Quick, one-time tasks such as summarizing a document, extracting a few values, reviewing an image, or analyzing a small set of files.	A maximum of five files can be uploaded per conversation, with a combined size limit of 50 MB across all uploaded files.Additionally, each individual multimedia file mustn't exceed 25 MB.	No, the default LLMs are used for image and multimedia processing.
Build reusable knowledge for retrieval	Document Tool	Guides, FAQs, handbooks, procedures, and other reference content that should be searchable and reusable by the agent over time.	A single file size limit of 25 MB.	No, the default LLMs are used for image and multimedia processing.
Process business object attachments in the background	Document Processor	Asynchronous processing of files attached to business objects, automated ingestion, workflow-driven document handling, scanned PDFs, image-heavy PDFs, long DOCX files, and dense text files.	A single file size limit of 25 MB.	No, LLM can only be configured when LLM-based data extraction with Document Schema is enabled. When it isn't enabled, the default LLMs are used for image and multimedia processing.
Extract structured fields from files	Document Processor with Document Schema	Deterministic field extraction, structured output, and cases where only specific values need to be extracted from a document.	A Document Schema is required. A single file size limit of 25 MB.	Yes, when LLM-based data extraction with Document Schema is enabled. For more information, see Add a Document Schema to a Document Processing Node of Workflows.

Best Practices

Follow these best practices to improve reliability, reduce processing failures, and choose the right file-handling approach for your use case.

• Keep files small and focused

  Large files can fail even when they are within the recommended size limits. File size is not the only constraint; extracted text, embedded images, OCR output, tables, and token expansion can also affect processing. Split large or dense files into smaller documents, especially long reports, large spreadsheets, scanned PDFs, and image-heavy PDFs.

• Use clear, readable source files

  Fully scanned PDFs and image-heavy PDFs can be harder to process. Use text-based PDFs when possible. For scanned documents, make sure pages are upright, readable, high contrast, and not blurry or skewed.

• Use chat uploads for lightweight tasks

  Use chat uploads for short, one-time tasks such as summarizing a small document, extracting a few values, reading an image, or reviewing a small set of attachments. Use another processing option when the content must be reused, searched, or processed as part of a background workflow.

• Use Document Tool for reusable knowledge

  Use Document Tool when content should be searchable and reusable over time, such as guides, FAQs, handbooks, procedures, and reference documentation. Keep source documents focused by topic so retrieval returns more relevant results.

• Use Document Processor for background processing and extraction

  Use Document Processor when files are attached to business objects, need asynchronous processing, or require structured extraction. For field extraction, provide a Document Schema and clear instructions so the processor returns only the required values.

• Avoid sending full extracted content to the LLM

  Extracted content can be much larger than the original file and may increase cost, slow processing, or exceed model context limits. Send only the relevant sections, fields, pages, or chunks needed for the task. For large files, use retrieval, filtering, summarization, or a document schema.

• Simplify structured files before upload

  Large spreadsheets, CSV, JSON, and XML files can contain hidden, unused, or deeply nested content that increases processing complexity. Remove unused columns, empty rows, hidden sheets, unnecessary formulas, formatting, and irrelevant records before processing.

• Review archive contents before upload

  ZIP file processing depends on the files inside the archive. Include only supported and relevant files, and avoid unnecessary nested folders, duplicate files, or archives that expand into very large content.

Estimated Content Capacity

The following estimates provide a general guideline for the amount of content that can typically be processed within a 25 MB file size limit. Actual capacity may vary depending on factors such as document complexity, formatting, embedded media, and file structure.

Estimators

File type	Estimated Capacity for a 25 MB file
PDF, text-based	300 to 1000+ pages.
PDF, image-heavy or scanned	50 to 200 pages.
DOCX	100 to 500+ pages.
PPTX	50 to 150 slides.
TXT / Markdown / HTML	5 to 15 million words.
CSV / JSON / XML	Highly variable and best kept smaller if dense or deeply structured.
XLSX	Tens of thousands to hundreds of thousands of cells, depending on formulas, formatting, and sheets.
Images	10 to 100 images, depending on resolution and compression.
Multimedia	About 15 to 60 minutes, depending on format and bit rate.
ZIP archives	Depends on the files inside and extracted content still needs to be processed.