DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Grants WSL2 GPU device permissions (/dev/dxg and /proc) to the DSH Landlock sandbox so CUDA inference works inside a confined agent session.
Reorganizes DSH session titles into short, consistent formats with template customization, local semantic icons, model-based generation and refinement, and a native rename workbench.
Turns DeepSeek Harness into a spec-driven Odoo development pipeline with a closed-loop on a real instance, covering module development and functional data configuration.
Image-generation plugin for DeepSeek Harness: configure multiple API gateway providers and models in settings, then generate and display images in conversation via the image_generate tool.
Boolean algebra toolbox for DeepSeek Harness: parse propositional-logic expressions to generate full truth tables, canonical DNF/CNF, NNF, and NAND/NOR gate networks with zero runtime dependencies.
A one-click button beside send that asks your model to diagnose gaps in your draft prompt, then offers three patch-style suggestions you can cherry-pick and merge back without overriding your send action.
Adds Token Market as an OpenAI-compatible model provider for DeepSeek Harness, giving one API key access to multiple model families with streaming and tool calling.
Auto-syncs the gpt-load model pool into DeepSeek Harness as a single provider group, with per-model OpenAI or Anthropic protocol selection and automatic capability mapping.
A DSH plugin for deepseek-v4-flash-vision-exp that splits large images into 800x800 lossless tiles for refined vision-model reading, offering smart, pipeline, and full strategies to overcome downscaling loss.
Resumes an interrupted streaming reply from its already-generated partial output instead of restarting from scratch when the endpoint times out or disconnects.
Integrates OpenCode Go as a first-class model inference provider in DeepSeek Harness with session routing, live catalog refresh, and multi-protocol dispatch.
Wraps Microsoft MarkItDown as a DSH tool so the model can read PDFs, Office documents, and more as Markdown, with four-tier engine auto-detection and a built-in fallback.