Blog
Releases, technical write-ups and events from the people building Docling.
NVIDIA RTX Spark brings the full CUDA AI stack to a pocket-sized device. Docling ships day-0 support so you can run local document understanding and agentic pipelines from the moment it lands.
Meet the Docling team in Atlanta October 26–29 — four hands-on labs and three talks on agentic RAG, event-driven ingestion, DocLang, and Docling for IBM watsonx.
On August 27, 2026, Docling became a Graduated project of the LF AI & Data Foundation, the highest stage in its lifecycle. We got here thanks to a community that grew from filing bug reports to opening more than half of all pull requests.
How Docling extends into agentic AI through two complementary interfaces — the Docling MCP server and Docling Skills, for AI clients like Claude, Codex, or IBM Bob.
IBM now offers Docling as a managed service. Here is what that means for the open-source project, plus the new batch endpoint and the lightweight Docling Service Client.
Introducing ChartNet, a CVPR 2026 million-scale multimodal dataset of aligned chart images, code, tables, summaries, and reasoning traces — the kind of supervision that powers chart understanding in models like Granite-Docling and Granite Vision.
Document layout analysis evaluation metric that works across heterogeneous class taxonomies
Meet the Docling team in Atlanta May 11–13 — four sessions covering scalable document AI, RAG on RHEL docs, agentic apps with Python, and large-scale processing with Ray Data on OpenShift AI.
How Docling's LaTeX backend turns source into structured documents without going through a PDF round-trip.
Accelerating Document Processing with NVIDIA Nemotron OCR
A detailed educational guide to how Docling extracts chart values, with end-to-end CLI/Python workflows, quality checks, and production integration patterns.
