Self-host Docling Studio

Docling Studio runs in the browser and lets you analyze PDF files. After uploading a document, you choose which Docling features to enable — optical character recognition, table structure recovery, code and math detection, image classification, and picture extraction. The viewer overlays color-coded boxes on each PDF page and keeps a side panel in sync with the current page. Extracted material can be cut into chunks using several strategies with token limits you control. If you enable the ingestion profile, chunks go to OpenSearch for vector and keyword search, and the entire document hierarchy is mirrored into Neo4j as a graph of sections, paragraphs, tables, pages, and chunks. The interface also handles document listing, deletion, search, and filtering by index status, keeps a history of past analyses, exports to Markdown or HTML, enforces configurable size and rate limits, and offers dark/light themes plus French and English. Under the hood a Vue 3 frontend (nginx on port 3000) talks to a FastAPI backend (port 8000) backed by SQLite. Two container images exist: a large one (~1.9 GB) that runs Docling locally on CPU, and a small one (~270 MB) that calls a separate Docling Serve instance. The codebase is MIT-licensed and sees regular updates.