DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
A dynamic prompt generator that assembles structured PRD, codebase analysis, and bug analysis prompts tailored to technology stacks and analysis dimensions.
Detect whether workers in an image wear reflective vests, returning bounding boxes, confidence scores, and safe/unsafe labels via URL, Base64, or file path input.
Chinese speech recognition plugin for DeepSeek Harness, powered by the Paraformer INT8 model, accepting audio URLs, Base64-encoded data, or local file paths.
A model-inference plugin that detects electric bicycles in images via URL, BASE64, or file path input, returning bounding boxes, confidence scores, and labels.
Recognize insect or arachnid species names and taxonomic classification (order, family, genus, species) from image links, BASE64 data, or file paths.
A math computation plugin for DSH covering expression evaluation, array statistics, matrix operations, financial calculations, and calculus powered by SymPy, Polars, and NumPy.
A PDF reading service for DeepSeek Harness that supports text extraction, page-range reading, metadata retrieval, and keyword search.
A plant recognition plugin for DeepSeek Harness that identifies plant names and taxonomic classification (family, genus, species, or subspecies) via image URL, base64 data, or file path.
An ID card OCR plugin that auto-detects front and back sides, extracts fields such as name, gender, address, and ID number, validates the ID number, and accepts image URLs, BASE64 strings, or file paths.
A YOLO26-based object detection plugin that detects 80 COCO classes in images, outputting bounding boxes, confidence scores, and class labels via URL, Base64, or file path input.
An OCR plugin for passport recognition that extracts passport number, names, gender, nationality, dates, and MRZ codes, accepting image URL, Base64, or file path as input.
Detect vehicle types (car/truck/bus/motorbike/tricycle/carplate) in an image and return bounding boxes, confidence scores, and labels, with URL, Base64, or file-path input.