DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
A SillyTavern-compatible worldbook injector that injects prompt snippets at request level via keyword, regex, or always-on triggers with zero context pollution, plus agent-driven lore lookup.
One-click Claude Pro/Max OAuth sign-in for DeepSeek Harness with automatic credential sync, live model discovery, and real-time quota tracking.
A self-contained safe message renderer for Harness with ordered regex, macros, constrained expressions, sanitized Markdown/HTML, and eight immersive document templates.
Configure reasoning tiers and vision capability for custom models in DSH settings, replacing manual settings.yaml edits with a simple form.
A unified human-brain-style context compaction system that composes seven community compression plugins into one coordinated panel with engine arbitration and brain_verify one-click checks.
Drives your logged-in Claude Code CLI as a first-class model inference provider for dsh, with no API key and no runtime dependencies.
Adds a Command Code panel to the DeepSeek Harness settings, showing credits, usage windows, billing details, and a syncable model catalog.
A private multimodal integration suite for DeepSeek Harness that unifies visual evidence, ComfyUI, Unity, Adobe, CAD, and Ren'Py workflows.
A model-aware reasoning strength slider for DSH with three visual modes, lightweight local preview, and no fabricated tiers.
A DeepSeek Harness adapter plugin for locally deployed Qwen models served via vLLM's OpenAI-compatible endpoint, with per-model multimodal switching and fully configurable reasoning effort levels.
A capability profile panel for DSH sessions, showing per-model session counts, tool usage, error rates, top error signatures, and multi-model comparison with zero runtime dependencies.
Visualizes prompt cache hit ratios per model call in DSH and diagnoses why uncached tokens changed between adjacent inputs.