DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Diagnose WSL2 NVIDIA GPU visibility, VRAM pressure, and inference port conflicts to prevent multi-engine OOM.
A DeepSeek Harness community plugin that integrates Doubao Search (Volcengine) as an independently-billed web search provider via ctx.web, with runtime-configurable settings.
Auto-fills model capability parameters such as context window, output limit, reasoning, and compat for custom providers in DeepSeek Harness by matching model IDs against the pi.dev catalog.
One-click or auto-sync of model lists across all DeepSeek Harness providers via their /v1/models endpoints, with replace semantics and per-provider failure isolation.
Turns vague requests into structured, production-ready prompts with auto-fill, context anchoring, and iterative revision—zero key management via DSH native model routing.
An OpenAI-completions-compatible adapter for custom gateways such as vLLM and LM Studio, with correct system-role handling, per-model thinking-format passthrough, and multi-image vision input.
Enhanced Command Code LLM provider plugin for DSH-Desktop with multi-account rotation, streaming generation, and a usage dashboard.
Multi-backend search-chain orchestrator for DeepSeek Harness with built-in Ollama, Exa, and DeepSeek adapters, supporting dynamic fallback and policy management.
Deterministic, recoverable tool-result compression, cache-aware 62.5% automatic compaction, and a Web conversation dashboard for DeepSeek Harness.
Image generation for DeepSeek Harness via OpenRouter's unified Image API, supporting prompt-to-image, reference-based editing, model aliases, capability gating, and inline chat display.
A DeepSeek Harness plugin that wraps tariff data standardization and descriptive statistics into auditable, reproducible pipelines with XLSX/CSV ingestion, quality checks, and grouped reporting.
Automatically tiles oversized images into up to 9 HD tiles with row/column coordinates for DeepSeek vision models, preserving fine text and chart details lost to auto-scaling.