DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Registers Zhipu GLM models as a provider route with dual billing-channel switching, local key management, usage statistics, and a Web settings card.
Select a different model and reasoning effort for DSH subagents without changing the main conversation model.
A dsh Web plugin that brings plugin health grades (S–D), version-drift and compatibility checks, plus one-click install/uninstall directly into the workspace.
Preview and download the exact request frame sent to the model, including system prompt, tool declarations, and the full message array, rebuilt from the session log.
A prompt design companion for Harness that optimizes or designs executable coding-agent prompts using project context, streaming results into the composer without auto-sending.
A zero-dependency local linear algebra toolbox for dsh covering matrix multiply, determinant, inverse, RREF, linear system solving, and vector operations with 10-decimal precision.
Keep DeepSeek Harness minimal by default, requesting browser, documents, and code navigation on demand and releasing them when tasks are done.
Auto-detects the real runtime context window of local llama.cpp or Ollama servers and bidirectionally syncs it into DSH settings, zero config.
A zero-dependency complex-number math toolbox for DeepSeek Harness providing parsing, arithmetic, polar conversion, principal-value functions, and n-th roots.
Maintain a custom context in DSH settings and inject it as a systemPrompt section into every model call, with persistent storage.
A DSH web plugin that adds a Model Capabilities page to the settings panel for graphically editing input modalities, context windows, output caps, and image budgets.
Adds context window, compaction threshold, and reasoning effort controls to pi-ai provider cards on DSH Web Settings, applied live without restart.