Independent guide — not affiliated with IBM or the Docling project. Every command verified on Docling 2.143.1.
  1. Home
  2. Try online
Live demo

Try Docling online — no install, in your browser

The quickest way to see Docling's AI at work is IBM's official Granite-Docling demo. It converts one page image at a time with the 258M-parameter vision-language model — a preview of the VLM pipeline, not the full standard pipeline.

Verified on Docling 2.143.1Updated 27 Oct 20264 min read
In short
  • The demo runs Granite-Docling-258M on a shared Hugging Face GPU (ZeroGPU): expect a queue at busy times.
  • It takes one page image (PNG/JPEG) — screenshot or export a single PDF page first.
  • Prompts it understands: “Convert this page to docling.”, “Convert this table to OTSL.”, “Convert formula to latex.”, “Convert code to text.”, “Convert chart to OTSL.”
  • For whole PDFs, Office files and the standard pipeline, run Docling locally — or start docling-serve's UI in two commands.
Granite-Docling-258M · official demo on Hugging Face (shared GPU)Open full screen ↗

The demo is hosted by IBM on Hugging Face. Loading it connects your browser to huggingface.co and hf.space.

Your first conversion in three steps#

  1. Prepare one page

    Screenshot or export a single page as PNG or JPEG. Straight, sharp captures work best; the demo also includes sample pages in English, Arabic and Chinese.

  2. Upload it or pick a sample

    Choose a document from the gallery or upload your image.

  3. Send a prompt

    Start with Convert this page to docling. — the model returns DocTags, which the demo renders with layout boxes.

Prompts that work#

PromptReturns
Convert this page to docling.The whole page as DocTags (layout, text, tables, pictures)
Convert this table to OTSL.Table structure in OTSL markup
Convert formula to latex.LaTeX for a formula crop
Convert code to text.Code from an image of code
Convert chart to OTSL.Chart values as a table
Extract all picture elements on the page.Picture locations
What is the Net income in 2008?A question about a table (sample document)

Prompts taken from the demo's own sample set. The model card describes multilingual support (Japanese, Arabic, Chinese) as experimental.

What the demo is — and isn't#

Online demoDocling on your machine
ModelGranite-Docling-258M (VLM)Standard pipeline by default, or any of 22 VLM presets
InputOne page image36 formats, whole documents, folders
OutputDocTags rendered in the pageMarkdown, JSON, HTML, chunks and 8 more
PrivacyImage is sent to Hugging FaceStays on your machine
CostFree, queuedFree (MIT), your hardware

Run the full Docling UI locally#

docling-serve includes a browser playground that converts real documents with all the options:

bash
pip install "docling-serve[ui]"
docling-serve run --enable-ui
# open http://127.0.0.1:5001/ui

Or with a container: docker run -p 5001:5001 -e DOCLING_SERVE_ENABLE_UI=1 quay.io/docling-project/docling-serve-cpu. Details in the docling-serve guide.

See the real pipeline on a real paper#

Want to know what the standard pipeline produces before installing? Our chunk visualizer shows the actual output of Docling 2.143.1 on its own technical report, and the benchmark shows its tables next to other tools'.

Frequently asked questions

Is there an online version of Docling?

There is no official hosted converter for whole documents, but IBM publishes a free Granite-Docling demo on Hugging Face that converts single page images. For full documents, install Docling locally or run docling-serve with its web UI. IBM also offers a managed 'Docling for IBM watsonx' service.

Is the Docling online demo free?

Yes. It is a public Hugging Face Space. It runs on shared GPUs, so you may wait in a queue.

Can I upload a PDF to the online demo?

The demo expects a single page image. Take a screenshot or export one page of your PDF as PNG or JPEG.

How do I get a local Docling web interface?

pip install "docling-serve[ui]" and run docling-serve run --enable-ui, then open http://127.0.0.1:5001/ui. It converts full documents with the options of the standard and VLM pipelines.

Sources