xby-detect

xby-detect turns your DeepSeek Harness chat into an on-demand vision workspace. Instead of spinning up separate model services and writing inference scripts for every task, you call one API key and get back detection boxes, species labels, or segmentation masks right inside the conversation. Twenty-plus vision capabilities are packaged into a single plugin, eliminating the friction of wiring up individual model endpoints.

The capability spread is broad. On the general side you get COCO-80 object detection, ImageNet-1000 classification, oriented bounding-box detection across 15 categories (airplanes, bridges, swimming pools, and more), 17-keypoint human pose estimation, and instance segmentation. Specialized verticals cover everyday objects, cellphones, e-bikes, fire, cigarettes, gestures, pedestrians, vehicle types (car, truck, bus, motorbike, tricycle, plate), safety helmets, reflective vests, and even dish recognition. In the biological domain it identifies insects down to order/family/genus/species, plants to species or subspecies, animals, birds, and wild animals, and can even classify a pet s facial expression into Angry, Happy, Relaxed, or Sad. A single general recognition endpoint tags over 50,000 object categories. Every capability accepts image URLs, Base64 strings, or local file paths, so it slots into whatever pipeline you already have.

It is built for developers, researchers, and security operators who want a quick look-and-verify workflow: photograph a leaf in the field and get a pest ID, scan a construction site photo for helmet and vest compliance, batch-check surveillance frames for fire or vehicle presence, or split a product photo into instance masks. One chat message, one tool call, immediate structured results.