DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
A context compression plugin for DeepSeek Harness that automatically generates summaries when approaching the context limit and handles overflow.
Unified model routing plugin aggregating models from multiple vendors, featuring tiered routing, automatic failover, and health-based selection.
DeepSeek Harness translation plugin using local LLM, no API keys needed. Supports draggable, pinable cards with auto language detection.
DeepSeek Harness plugin splitting agent review into three independent skills (session, repo, assets), zero dependencies.
A DeepSeek Harness medical image analysis plugin based on the MedOmni strategy, providing modality-specific tools for reporting, localization, segmentation, and classification.
One-click LLM-powered polishing for drafts, refining text into clearer content with results filled back into the input box.
A token consumption statistics plugin for DeepSeek Harness Web UI that visually displays daily and total usage data through charts.
dsh-codex-relay is a DeepSeek Harness plugin that integrates Codex as a model inference backend, delegating requests to Codex's native agent loop for tool calls and command execution.
Expose DeepSeek Harness as an OpenAI-compatible API server, with real agent sessions backing each API call for chat completions and model listing.
dsh-loghud is a local development error monitoring plugin for DeepSeek Harness, supporting error identification in multiple languages with manual AI diagnosis.
Provides read-only model request concurrency monitoring for DeepSeek Harness Web, displaying active requests, peak concurrency, and supplier statistics without intercepting or modifying requests.
dsh-auto-vision enables models without image input support to process screenshots by auto-recognizing with a vision model and adding descriptions as folded context entries, preserving the original message.