Self-host DeepSeek-OCR-WebUI

DeepSeek-OCR-WebUI wraps the DeepSeek-OCR-2 vision model in a browser-based document recognition app with a REST API. Uploaded images and PDFs can be processed in seven modes: document-to-Markdown, general OCR, plain text, chart and formula parsing, image description, find-and-locate with bounding-box annotation, and a custom prompt mode. The Vue 3 interface adds a page sidebar with thumbnails and reordering, a real-time processing queue, health monitoring, and a four-language UI (English, Simplified and Traditional Chinese, Japanese). The FastAPI backend exposes /ocr, /ocr-pdf, /pdf-to-images and /health endpoints, and a bundled Model Context Protocol server lets AI assistants call the OCR service. To self-host, run the all-in-one Docker image (GPU recommended; the model is pre-downloaded inside the image) or install natively on Linux or Apple Silicon Macs with conda/pip and start.sh. CPU-only Docker is supported but described as very slow; the docs list a minimum of a 16GB+ VRAM NVIDIA GPU, 16GB RAM, 50GB disk and CUDA 11.8+.