DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
A math computation engine for AI assistants, supporting expression evaluation, matrix operations, unit conversion, number theory analysis, and session management.
Detect all cellphones in an image via URL, BASE64, or file path, returning bounding boxes, confidence scores, and labels.
A helmet, head, and person detection plugin that accepts image URLs, BASE64 encodings, or file paths and returns bounding boxes, confidence scores, and labels for all detected targets.
A Model Context Protocol server for AntV visualization libraries, providing intelligent topic detection, requirement parsing, and official documentation retrieval.
Recognizes handwriting text from images via URL, Base64, or file path, automatically detecting and extracting content from notes, signatures, and handwritten forms.
Generate high-quality AWS architecture diagrams, Kubernetes diagrams, sequence diagrams, and more in GitHub Codespace via Amazon Q CLI and MCP server.
MCP-based numeric calculator plugin supporting addition, subtraction, multiplication, division, power, square root, and factorial with persistent API key management.
Detects people, pets, vehicles, flames, and cardboard boxes in an image, returning bounding boxes, confidence scores, and labels via URL, Base64, or file-path input.
A cognitive-enhancement plugin for DeepSeek Harness that adds structured dialectical reasoning, enabling claim decomposition, thesis-antithesis-synthesis progression, and argument strength evaluation.
Adds a persistent output style manager to DeepSeek Harness Web Settings, with built-in and custom Markdown styles that take effect from the next model step without restart.
A zero-dependency, local-first MCP server that extracts structured event-research fields (type, date, amount, shares) from A-share disclosure text, with no network or API key required.
Per-task automatic model hot-switching for dsh: routes image tasks to a vision model while keeping the default for everything else, zero extra tokens, one-click enable.