DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Local, deterministic per-turn model router for DeepSeek Harness that auto-switches between economy and quality models using keywords and input length, with zero extra token cost.
Safe unattended task continuation plugin with deterministic network recovery, unique recommendation selection, binary confirm, and Plan Mode exit; all capabilities are off by default.
Unified provider authentication, model catalog, and session routing for DeepSeek Harness with built-in Qwen Code and Codex support.
Aggressive local-first context compaction for DeepSeek Harness agents that shrinks conversations to fit small llama.cpp context windows with zero API cost and no data egress.
One-click prompt enhancement preserving original intent, with 22 thinking methods, a local vault, and zero telemetry.
Aggregates OpenRouter free model ratings, TeamoRouter pricing and balance, live availability probes, and usage stats into one card with one-click model switching.
A one-stop plugin for checking AI subscription quotas and balances across Codex, Kimi, DeepSeek, 302.AI, and OpenCode Go, with auto-refresh and in-session indicator.
Analyzes branch diff and auto-generates a Conventional Commits PR title, five-section description, and self-review checklist; optionally opens a PR via gh CLI.
A relay enhancement plugin for DeepSeek Harness that autofills reasoning efforts, syncs relay models, and enables a dynamic Fast priority tier.
Adds Codex as a native conversation backend to the official DeepSeek Harness Web UI, preserving workspace, history, and approvals without switching interfaces.
Turns named subagent profiles into model-visible delegation tools with selectable backends, custom or inherited routing, and per-profile reasoning policies.
Powers DeepSeek Harness with MinerU-based PDF parsing, offering layout analysis, formula and table extraction, and asynchronous jobs with smart caching.