Label Studio becomes the first annotation platform with native DocLang export, adding a purpose-built Docling interface and a live connection to Docling models for IBM watsonx
San Francisco August 25, 2026
HumanSignal, the creator of open source Label Studio and provider of human data services for AI teams, today announced an integration with Docling, IBM's open source toolkit for converting PDFs, scans, and office files into structured, AI-ready data. The integration connects the two projects directly, giving teams a way to create ground truth for their own documents, measure how well Docling performs on them, correct its output, and build the labeled data needed to fine-tune models for custom formats.
Docling handles a wide range of documents well, but most organizations rely on internal formats no public model has seen: lab reports, claim forms, invoices, engineering drawings. Without ground truth, teams can't measure conversion quality at scale, track whether a new model version improved or regressed, or produce the structured examples fine-tuning requires. Label Studio closes that gap with a visual review and annotation workflow built for the DocLang format.
The integration includes three components:
A new DocLang interface in Label Studio. Co-developed by the Label Studio and Docling teams, the interface lets annotators work directly on the document, drawing bounding boxes around element types (with optional OCR powered by Tesseract) and defining reading order. It supports ordered and unordered lists, merged regions, and detailed table structures. A sidebar shows the live DocLang structure as the document is annotated, in both XML and rendered form, and the DocLang XML is saved alongside the annotations for downstream use.
Native DocLang export. Label Studio is the first annotation platform to export DocLang natively. Teams can download annotated data as a .dclx DocLang Archive file containing the structural data and related page images, ready for evaluation harnesses, fine-tuning pipelines, or any tool that reads the standard.
A live connection to Docling models. A new Label Studio ML Backend for Docling for IBM watsonx runs documents through hosted Docling models in real time. Predictions appear in Label Studio automatically, so reviewers can correct them and promote them to ground truth, which is significantly faster than annotating from scratch.
Together, these support three common workflows: building gold-standard sets to evaluate Docling on an organization's own documents, routing low-confidence or high-stakes conversions to human review in production, and creating correctly structured training examples for fine-tuning.
The partnership builds on HumanSignal's role as a founding member of the DocLang working group. Earlier this year, IBM, ABBYY, NVIDIA, Red Hat, and HumanSignal launched DocLang, an open standard under the Linux Foundation that defines how documents should be represented for AI systems.
Availability
The DocLang interface is available now in Label Studio Starter Cloud and Label Studio Enterprise. Compare Label Studio editions and sign up for a free two-week trial.
The Docling ML Backend is available today in beta on GitHub.
New Docling for IBM watsonx accounts include a 30-day free trial with 5,000 pages