DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Renders long text messages into 800x800 images for vision models, reducing Chinese input tokens by ~2-2.5x while keeping all content intact.
A unified image studio plugin integrating four 88api.ai models for text-to-image, multi-reference editing, 2K/4K output, and sequential batch generation with a single install and key.
A DSH plugin that bridges third-party LLMs via OpenAI Responses or Gemini protocols, offering hosted search, model discovery, and Code Mode support without modifying the host.
An all-in-one visual creation toolchain for DeepSeek Harness: prompt reverse-engineering, deterministic palette extraction, prompt auditing, and multi-provider image generation in a single plugin.
Adds a Model Health panel to settings that lists all configured models and supports one-click batch testing of availability and latency.
A card-based plugin manager for DSH with toggle, detail view, one-click update, search, and real-time error display.
Routes Claude, OpenAI/Codex, and Gemini via Antigravity OAuth into DeepSeek Harness through a single local CLIProxyAPI instance.
A vibe-learning plugin for DeepSeek Harness enabling collaborative outlines, interactive lessons, FSRS-based review scheduling, quizzes, and in-session note-taking.
Reuses the official Grok CLI OAuth login to provide a subscription-based Grok inference route, dual-mode auth panel, and live model discovery for DeepSeek Harness.
Local offline voice input plugin for DSH Web GUI, powered by FunASR and SenseVoiceSmall for real-time dictation with automatic text insertion.
Routes DeepSeek Harness sub-agents to local llama.cpp servers via the OpenAI Responses protocol, reducing API costs.
A reasoning intensity slider for DSH that surfaces only model-declared levels, offers native and energy visual modes, and submits locally with zero fabricated options.