Parse multilingual documents for agent ingestion with dots.ocr
Use dots.ocr when an agent needs repeatable multilingual document layout parsing before retrieval, extraction, review, or downstream automation.
npx skills add agentskillexchange/skills --skill parse-multilingual-documents-for-agent-ingestion-with-dots-ocr
Use dots.ocr when a workflow needs to convert PDFs, document images, web screens, charts, diagrams, or scene text into structured outputs an agent can inspect and reuse. The operator sets up the dots.mocr model through vLLM or Hugging Face, runs the parser against images or PDFs, chooses the prompt mode for full layout, OCR-only, web parsing, scene spotting, or SVG-oriented output, then feeds the resulting structured text or visual markup into retrieval, extraction, QA, or review workflows. Invoke this instead of using the demo or normal manual OCR when document parsing must be repeatable, local or self-hosted, and attached to an agent pipeline with reviewable artifacts. The scope boundary is document and visual layout parsing for agent ingestion; it is not a general OCR product card, document management system, or broad VLM framework listing.
What this skill actually does
Inputs and prerequisites: Python, vLLM or Hugging Face Transformers, dots.mocr model weights.
Setup notes: Clone https://github.com/rednote-hilab/dots.ocr, create the documented Python 3.12 environment, install PyTorch for the target CUDA runtime, run `pip install -e .`, download model weights with `python3 tools/download_model.py`, then deploy through vLLM or run parser commands such as `python3 dots_mocr/parser.py demo/demo_pdf1.pdf –num_thread 64`.
Source and verification boundary: use https://github.com/rednote-hilab/dots.ocr as the canonical reference before running the workflow; keep commands, API calls, CLI usage, and generated outputs reviewable against that upstream source.
Framework fit: publish this as a Multi-Framework workflow only when the operator can invoke the documented toolchain directly, rather than treating the upstream project as a generic product listing.