DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
A pure Node.js plugin for computing correlation coefficients with no network dependencies, designed for data relevance analysis in model inference scenarios.
A pure Node.js combination calculation tool with no network dependencies.
Basic training plugin that returns clear suggestions based on input parameters, built in pure Node with no network dependencies.
A local Node.js plugin for DeepSeek that provides tools to list and query movie genre entries by id or name, with no network dependencies.
A pure Node.js model inference plugin that runs entirely locally with no network dependencies.
A session-scoped LLM request and response inspector for DeepSeek Harness Web, providing call snapshots, semantic API comparison, and state filtering without altering model I/O.
Transcribes every image in the agent stream into detailed text via a multimodal understander so text-only models can handle visual tasks, with automatic recovery on image-related failures.
Fixes tool-call retry failures caused by spurious sandbox permission params from GPT-like models and adds a skip button for stuck calls.
Integrates ZhiPu GLM vision models into DeepSeek Harness and provides a glm_vision tool so text models can perceive images through GLM.
A private long-term companion and reflection workspace for DeepSeek Harness featuring governed memory, a 3D constellation, photo-particle scenes, and nine reflection spaces where the user owns every conclusion.
A third-party channel plugin that connects a Harness Agent Preset to Narra Messenger, supporting streamed text replies and voice-instruction mode.
Fixes streamed tool-call identity loss, auto-corrects common gateway 400s with one retry, and provides custom Chat Completions routes for DSH.