DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
An LLM adapter for DeepSeek Harness that connects Kimi models through three routes covering Code subscriptions and CN/global platforms, with thinking, tool calling, and multimodal image support.
Local GPU background-removal plugin for DSH: one-click cutout in chat, three swappable model tiers, no API key, fully offline.
Detects long reasoning chains in real time, segments them semantically, and produces per-segment summaries with optional small-model refinement, displayed live in the chat area with persistent history.
Automatically bridges vision to text-only models by calling a vision model to describe images and feeding the result back, while leaving native vision models untouched.
Publish agent-generated text, Markdown, HTML, and documents as ShareOne share links with password protection, watermarks, custom slugs, and comment collaboration.
Save and one-click switch frequently used model and reasoning-effort combinations in the DeepSeek Harness model menu, with Alt+1-9 keyboard shortcuts and browser-local persistence.
A DSH plugin adding model reasoning-level injection, CLI request mimicry, message timeline and collapse, auto history loading, global instructions, and independent module toggles for daily optimization.
Injects DSclaw persona files (SOUL, IDENTITY, AGENTS, TOOLS) into the session system prompt for agent identity governance without intercepting tool calls.
LLM cost-tracking plugin for DeepSeek Harness that auto-records token usage and cost per API call with peak/off-peak pricing, subscription quota monitoring, and a local visual dashboard.
Learning-based agentic router that routes requests to the optimal model tier by task complexity, powered by four expert recommenders, EXP3 meta-selection, and reward-driven continuous improvement.
Describe a scene in chat; the AI designs NovelAI V4.5 prompts and generates images, with tag lookup and reverse-prompt support.
A floating whale-elf on your DSH page that opens a lightweight draggable chat on click, streaming from the session model or any OpenAI-compatible endpoint without writing to session history.