DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Adds an image_analyze tool that sends local images or URLs to an OpenAI-compatible vision model and returns text descriptions, letting text-only models interpret visuals.
Create named model aliases in DeepSeek Harness to switch provider, model, and reasoning effort with a single click beside the input box.
A pure Node.js plugin that returns structured weekly report writing methods based on parameters, with no network dependencies.
A pure Node.js plugin that returns home storage plans based on parameters, with no network dependencies.
A dsh plugin that polishes, completes, and structures your prompt before sending, with multi-round rewriting, version rollback, and custom API support.
Integrates Claude as a selectable LLM provider in DeepSeek Harness via Anthropic's official Claude Agent SDK, preserving DSH's prompts and tools across all modes.
Local OCR plugin that recognizes handwritten Chinese notes and math formulas into Markdown text and LaTeX, fully offline with no data leaving your machine.
Adds vision capabilities to text-only models by reusing DSH providers or custom OpenAI-compatible endpoints, with auto image paste handling and a visual settings page.
Adds TokenLab as a model provider to DeepSeek Harness, exposing 134 models, 80 developer API tools, and async multimodal generation.
Turns model selection into intelligent routing: declare a virtual model ID bound to a candidate pool with priority or round-robin algorithms and automatic failover, transparently dispatching each call to a real model.
Prompt you to choose the provider/model driving each spawned subagent; nested subagents inherit the choice automatically. Zero dependency, built on the official LlmCallConfig waterfall hook.
A DeepSeek Harness plugin bundle providing token input preview with usage badges, cross-session token consumption stats, and one-click prompt optimization.