DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
The better-model-provider plugin enables per-model capability editing for DeepSeek Harness, including reasoning effort levels, input modalities, and token capacities.
Enables direct use of locally logged-in Codex in DeepSeek Harness without API key configuration, supporting image input and generation result callback.
dsh-codex-provider is a DeepSeek Harness plugin for OpenAI Codex, providing device code OAuth login, automatic token refresh, and provider management to access Codex models with ChatGPT subscriptions.
Routes eligible shell commands through rtk to compress output and save tokens without changing execution semantics.
Model usage plugin for DeepSeek Harness that tracks token consumption, estimates costs, displays account balance, and supports peak-valley pricing and currency conversion.
dsh-deepseek-vision is a vision-language gateway plugin that enables DeepSeek Harness to process images by routing them through a configurable vision model for non-vision-native models.
dsh-vision adds vision capabilities to DeepSeek, enabling zero-cost image recognition via channels like Doubao Web.
dsh-llm-verifier is a verifier plugin for DeepSeek Harness that uses an independent model to evaluate tasks and compare solutions, providing tools for comparison, selection, tracking, and verification.
A quality-of-life plugin for running Qwen3.8 locally in DeepSeek Harness, supporting llama.cpp, NInfer, and TabbyAPI backends with per-request thinking budgets and a live settings panel.
A lightweight DeepSeek Harness adapter for Unreal Engine 5.8's official MCP server, providing toolset discovery and invocation, UE source-context awareness, and read-only connection diagnostics.
DSH Director Toolkit is a DeepSeek Harness plugin that turns rough 3D ideas into comprehensive direction packs for software like Blender, Three.js, Houdini, or C4D.
Moves DeepSeek Harness file, subprocess, Bash, terminal, and LSP operations into a Tensorlake microVM sandbox.