DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
A persistent delayed-decision plugin for DSH Agents: resume the original session at a future point, gather fresh evidence, and reassess before acting.
Enable local visual perception for Agents: canvas rendering, clipboard capture, offline OCR, and pixel analysis without a vision model. Windows only.
An AI teacher agent preset that masters PDF documents, explains step-by-step as objective truth, and applies Socratic guidance with on-demand teaching skills.
A pure-config patch that routes dsh model inference through the local OpenLLM gateway and registers its MCP tools, no API key required.
Brings OpenCode Zen free models into DeepSeek Harness with zero configuration, no API key, and dynamic model discovery at runtime.
Aggregates free-tier upstreams for dsh with multi-key rotation, automatic failover, and a local OpenAI-compatible endpoint.
A curated 29-rule skill for writing AI video prompts, covering asset consistency, scene layout, character performance, and shot-by-shot generation, auto-loaded by the model or invoked manually.
Converts Office documents and PDFs (including scanned pages) to structurally-aware Markdown via MarkItDown, with OCR routing, available as CLI and agent tool.
A Claude Desktop-style model and reasoning-effort switcher for DSH Web: pick a model from the dropdown and tune inference strength with a slider, right beside the input box.
Per-provider concurrency gate for DeepSeek Harness model requests that queues surplus requests in a FIFO buffer to prevent HTTP timeouts.
Connects to your existing IndexTTS 2.5 or GPT-SoVITS installation to synthesize assistant replies sentence-by-sentence into speech by launching only the local worker process, with no WebUI required.
Explicit model routing for Main, Subagents, Web Search, image generation, and Plan execution in DeepSeek Harness; invalid routes fail explicitly with no silent fallback.