DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
This plugin automatically injects reasoning effort settings and necessary parameters for custom model providers, enabling selection in the composer model selector.
A playable world-model toolkit for DeepSeek Harness that enables strip inspection, route identification, run scoring, and research loop iteration.
This DeepSeek Harness plugin provides a muapi_generate tool for unified access to over 100 image, video, audio, and 3D generation models through MuApi.
dsh-argp is a context compression engine based on atomic reference graph pruning, achieving zero LLM calls through logical chain decision-making for selective forgetting.
A text adventure preset for DeepSeek Harness, offering nineteen world entries where the model outputs only novel text for immersive story creation.
dsh-llm-codex plugin integrates ChatGPT Codex with DeepSeek Harness, supporting OAuth login, model catalog, optional search and image generation, and remote settings management.
A DSH model proxy plugin that routes LLM requests by target domain, sending selected models through a proxy while others connect directly, with automatic retry on failure.
dsh-model-pro is a model management plugin for DeepSeek Harness that offers full lifecycle management of providers, smart routing, connectivity testing, and secure data uninstallation.
This plugin enables DeepSeek Harness to connect to the Google Anti Gravity Coding Plan and use top-tier AI models for seamless web-based chat and coding.
The dsh-thinking-levels plugin provides per-round thinking-level control for DeepSeek Harness, allowing automatic or manual adjustment of reasoning effort to optimize performance and efficiency in multi-step tool chains.
An embedded C/C++ firmware development toolbox with 4 agents and 8 skills covering FreeRTOS, interrupts, Keil MDK builds, and more, compatible across AI coding platforms.
The dsh-tool-json plugin provides secure and efficient JSON structured querying with JMESPath-inspired syntax, eliminating process overhead for agent handling of JSON data.