How to use the output#
CLI
Paste into a terminal where Docling is installed. Every flag is listed in the CLI reference with its default.
Python
A complete script: imports, PdfPipelineOptions, converter and export. See the Python API guide for what each object does.
docling-serve
A curl request for a running docling-serve: JSON for URLs, multipart form for local files.
Sensible starting points#
| Your documents | Start with |
|---|---|
| Reports, papers, contracts exported from Word/LaTeX | Standard pipeline, OCR off, tables accurate |
| Scans, faxes, phone photos | Standard pipeline, full-page OCR, the right OCR language |
| Mixed: digital PDFs with screenshots | Defaults (OCR on bitmap regions only) |
| Math-heavy papers | Formula enrichment, or the VLM pipeline with Granite-Docling on a GPU |
| Slides and spreadsheets | Nothing to configure: Office formats are parsed directly |
| Meetings, podcasts, videos | ASR pipeline (needs docling[asr] and FFmpeg) |
Frequently asked questions
Are the generated commands tested?
The option names, engine names, presets and defaults are read from Docling 2.148.0, and we ran representative combinations (default, OCR off, fast tables, engine and language choices, VLM and ASR set-up) on that version. Exotic combinations may still need an extra dependency, which the notes under the output point out.
Why does the CLI default to embedded images but the builder suggests placeholder?
The docling CLI embeds images as base64 by default, which made our 9-page test 17 times larger. The builder starts from placeholder because it suits RAG and LLM use; switch to embedded or referenced when you need the pictures.
Does the builder send my settings anywhere?
No. It runs entirely in your browser.