DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Configure custom provider routes and static model catalogs in DSH via Web UI or settings.yaml, supporting OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages.
On-demand image generation for DSH sessions with text-to-image and image-to-image support via any OpenAI-compatible endpoint, rendered inline as clickable thumbnails.
A high-fidelity context compaction backend for DeepSeek Harness that replaces basic summarization with faithful value preservation, session-language matching, and optional recursive chunking.
An LLM adapter plugin for Alibaba Cloud Bailian Kimi models in DSH, supporting multimodal image input, streaming reasoning chains, function calling, and token stats with zero build steps.
AI skill for Chinese official documents (38 types) with template-guided writing, audience profiling, GB/T 9704 formatting, and a no-fabrication policy.
An LLM adapter that bridges Devin's Connect RPC API into dsh's provider system, with streaming text, reasoning, tool calls, image input, and a dynamic model catalog.
Connects DeepSeek Harness to Zhipu AI GLM Coding Plan, unlocking the 1M-token context model glm-5.3 for large-scale code analysis and long-document processing.
An aggregate bundle that installs three math tools: Math.js numeric evaluation, Wolfram symbolic computation, and SymPy differential geometry.
Personal DSH workbench: custom banner, confetti celebration, foreign language micro-learning, task board with agent execution & cron, and one-click workspace openers. All local, no backend.
Adds native multimodal image paste, session image clearing, auto reasoning-effort injection, and Figma MCP bridging (design-to-code and AI canvas design) to DSH.
Use your locally installed Claude Code CLI as an LLM provider in DeepSeek Harness, with streaming output and local credentials — no extra API key required.
Convert CSV, Excel, and JSON data into interactive ECharts HTML with 26 chart types, multi-file merging, LLM-powered data transformation, and fully offline rendering.