Docling Document Parsing on Docker
A Docker image for a self-hosted document-parsing server, powered by IBM Docling, that converts PDF, DOCX, PPTX, HTML and more to Markdown or JSON with optional GPU acceleration.
This Docker image runs a self-hosted document-parsing server powered by IBM Docling, converting PDF, DOCX, PPTX, XLSX, HTML, Markdown and LaTeX documents into structured Markdown, JSON or HTML, with high-accuracy layout analysis, OCR and table-structure recognition. It exposes a document-to-text conversion API with synchronous and asynchronous modes, file upload and URL-based input, and chunking endpoints (hierarchical and hybrid) for feeding results into RAG applications.
It supports NVIDIA GPU (CUDA) acceleration through a separate image tag, an offline or air-gapped mode using pre-cached models, an optional web UI playground, and a helper script for model management; images are multi-arch (amd64 and arm64) and built and published automatically. It is part of a broader self-hosted AI stack alongside related images for Whisper, embeddings, LiteLLM, Ollama and an MCP gateway.
It is distributed as a shell-scripted Docker build.