DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Auto-selects and explains LLM route decisions with conservative failover before first visible output, serving purely as a routing-policy layer for DeepSeek Harness.
A vision bridge for text-only models that registers an analyze_image tool to answer image questions via any OpenAI-compatible endpoint.
A macOS desktop floating orb that projects DSH Agent status onto your screen in real time with breathing color changes and comic-style bubbles—no Web UI required.
Exposes a DeepSeek Harness instance as a SlashX v1 gateway endpoint with secure media I/O and rich-component delivery.
A minimal DeepSeek Harness plugin that registers a hello tool for name-based greetings, serving as the smallest complete dsh-plugin bundle example.
MCP adapter for DeepSeek Harness featuring a single proxy tool with lazy server start and on-demand search, describe, and call to save context window.
Enables image understanding for non-vision DSH models: accepts images in chat, auto-converts them to text via any OpenAI-compatible vision API, and provides a view_image tool.
A DeepSeek Harness plugin that gives text-only models vision capabilities via API, plus key-free local tools for crop, pixel-diff, OCR, and vectorization.
Enables text-only models to understand images by calling a vision model to generate descriptions via the describe_image tool.
Conditionally appends one system-prompt section by model route: file-discipline guidance for GPT, thought-language guidance for DeepSeek, zero overhead otherwise.
Gives text-only models image understanding by rewriting pasted images into text references and registering an analyze_image tool that routes to a vision model.
Integrate your ChatGPT/Codex subscription into dsh via OAuth device-code login, with provider-native compaction, multimodal image input, and usage tracking.