Download
DSH Vision Toolkit is an open-source DeepSeek Harness plugin that adds image understanding and visual automation capabilities to text-only AI agents. It integrates Agent Vision Toolkit directly into DeepSeek Harness Web and headless execution profiles. Users can paste images into conversations and ask questions without manually changing models or copying file paths. The software provides structured tools for image analysis, OCR, object detection, visual grounding, and screenshot comparison. Additional utilities support image cropping, foreground extraction, color analysis, and HTML screenshots. Its bundled agent skill defines workflows for interpreting images, rebuilding interfaces, and verifying visual results. The plugin includes configurable vision providers, managed runtime environments, and previewable artifacts. It offers integrated settings and automated image routing.
Features
- Native DeepSeek Harness Vision Integration
- Automatic Image Paste and Model Routing
- Structured Image Analysis and OCR Tools
- Visual Grounding and Pixel Difference Detection
- Screenshot-Based UI Reconstruction
- Vision Provider Settings and Artifact Previews
Project Samples
