Use Zerox when an agent workflow needs document OCR that preserves enough layout context for AI ingestion. The operator provides a PDF, DOCX, image, or remote file URL, configures model credentials, selects page and concurrency options, and receives Markdown or structured extraction output that can be reviewed before downstream automation. Operators can run it on local files, choose provider-specific models, extract only selected pages, preserve table-heavy formatting, save outputs for review, and pass normalized Markdown to retrieval or data-entry workflows. This is not a generic OCR library listing: the skill boundary is the document-to-agent-input workflow, including file conversion, vision-model extraction, optional schema extraction, and handoff to review or retrieval steps. Invoke it instead of using a hosted OCR product manually when the workflow must run from code, choose model providers, preserve table-heavy formatting, or feed normalized content into an agent pipeline. Keep the scope limited to extraction and review, not full document management.