DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
A read-only, streaming Excel analysis toolkit for dsh that profiles, filters, and pivots xlsx files with bounded memory and zero cell writes.
Enables multimodal vision for text-only DeepSeek models by auto-converting images into structured descriptions and providing active image viewing and OCR transcription tools.
Adds a native polish button to the DSH input box that refines your prompt using the current session model, with preview, conflict detection, and safe apply.
Converts PDF, Word, Excel, PPT and other files to Markdown locally via Microsoft MarkItDown before passing them to the model.
Enables text-only models to read images by delegating to a standalone vision model, returning plain-text descriptions without image bytes entering the main context.
A text-to-image plugin powered by the MiniMax image-01 model that generates PNG images across 8 aspect ratios directly within DSH conversations.
Injects WSL path mapping and shell facts into the system prompt so the model prefers Linux paths, handles /mnt/c and CRLF correctly, and respects Node 24 proxy settings.
Configure a dedicated model, reasoning effort, and persona profile for DSH subagents so delegated tasks run on a cheaper model while the main agent keeps its flagship.
Offline image analysis plugin for dsh powered by Apple Vision, providing OCR, image classification, face detection, and document layout parsing locally with no API key.
Configure the maximum number of automatic model-request retries (0-10) directly from DSH Desktop Settings, with no DSH core patches required.
Bridges all four z.ai GLM Coding Plan MCP servers (vision, web reader, web search, zread) as native DSH tools with zero-config credential reuse.
A local-first AI teaching workspace for school and tutoring teachers, covering lesson-prep templates, assessment diagnostics, and a traceable next-lesson improvement loop.