DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
ds-vision-plugin adds image input support to text-only DeepSeek models by automatically converting pasted images to text via a race among four vision models.
This plugin adds visual capabilities to text-only models in DeepSeek Harness by using free Gemini and GLM vision APIs for image description, OCR, and visual Q&A.
dsh-llm-fallback is a DeepSeek Harness plugin that provides automatic fallback to alternative providers when the primary one fails, ensuring uninterrupted model inference.
The dsh-image-bridge plugin provides image bridging capabilities for pure text models, enabling them to view images by injecting image paths as text marks for autonomous discovery of visual skills.
A plugin that equips DeepSeek Harness with multimodal capabilities via MiniMax, enabling image description, content generation, and voice interaction.
Registers a local Codex App Server as a main model provider for DeepSeek Harness, enabling efficient model inference.
This adapter converts the DeepSeek web version into a free model provider for DSH, enabling API key-free model inference via a local gateway with support for conversations, tool calls, and concurrent processing.
A full-duplex voice plugin for DeepSeek Harness with streamed ASR via SenseVoice, LLM, and TTS, featuring press-to-talk, live captions, and barge-in support.
The dsh-weneed plugin injects the 'we need' mode's chain-of-thought guidance into DeepSeek Harness agents, enabling executable and first-person reasoning with tool, command, and preset options.
A DSH performance plugin that merges adjacent text increments in model streaming output to reduce session event volume, alleviating lag and resource usage in long sessions and multi-agent concurrency.
Real-time chain-of-thought trajectory profiling for DeepSeek Harness, analyzing reasoning streams with keyword indicators and baseline classification, including optional session recording.
A self-model plugin for DeepSeek Harness based on Yogācāra philosophy, mapping consciousnesses and mental factors to track and regulate the agent's internal state.