dsh-shift-router：DSH 的两级模型路由插件
`dsh-shift-router` 是 DSH 的一个两级模型路由插件，旨在智能分配日常与关键轮次的模型使用。它通过 LLM Judge 将请求分类为 `fast` 或 `smart` 层级，采用“即时升级、趋势降级”策略，并在 Prompt Cache 热时暂缓切换以优化成本。该插件支持运行时故障转移，遇到服务异常时实行指数退避冷却，仅在同层内切换健康模型；同时具备任务级编排能力，由强模型规划并委派给弱模型执行，且强制任务上限。此外，它提供按层级统计的 token 与成本遥测。安装需 Node `>=22.0.0`，配置后可通过 GUI 或命令调整路由模式（auto/manual/off）
Read Moredsh-model-router：给 DSH 增加虚拟模型提供商并按激活提供商路由请求
`dsh-model-router` 是 DSH 的社区插件，旨在解决多提供商和多模型配置下的管理痛点。它通过注册虚拟提供商 `model-router`，将分散在各提供商的模型统一汇聚到单一选择器中，支持按模型 ID 严格归并。用户可在模型选择菜单或设置卡片中查看每个模型的可用提供商，并支持一键切换当前激活的提供商及拖拽调整优先级顺序。 该插件采用实时路由机制，会话中始终保存虚拟提供商标识，仅在每次 LLM 请求发出时解析并委派给当前激活的真实提供商，确保主代理、子代理及工具调用使用同一路由逻辑。插件具备自动同步能力，能随提供商或模型的新增删除自动清理配置并回落激活状态。它适用于 Web、C
Read Moredsh-subagent-model-router：为 DeepSeek Harness 子代理提供可选模型与路由配置
`dsh-subagent-model-router` 是一个为 DeepSeek Harness (DSH) 设计的插件，旨在统一管理子代理的模型路由、委派及后台结果汇合。该插件解决了不同任务（如检索、推理、审查）对模型能力要求不同导致的配置分散问题。核心功能包括：`subagent_model` 工具限制模型选择为已配置别名；`wait-for-subagents` 工具用于等待后台子代理终态结果；通过系统提示嵌入模型标签与路由说明以辅助编排器决策；提供只读模型目录视图及命名空间作用域的配置工具。此外，它支持 Web 设置页手动管理路由，并在 `~/.dsh/settings.yaml`
Read Moredsh-auto-model-router：为 DeepSeek Harness 提供 L0–L3 模型路由
dsh-auto-model-router 是 DeepSeek Harness (DSH) 的第三方模型路由插件，旨在解决不同请求对模型能力需求差异大导致的成本与效果平衡问题。该插件通过注册虚拟 `auto` provider，将请求智能路由至 L0–L3 四个能力层级，支持 heuristic lock、固定层级及 fully-auto 三种用户模式。其核心功能包括基于文本宽度的零延迟启发式锁定、可配置权重的关键词评分、可选 LLM 分类器、失败回退链以及成本优先、质量优先或平衡型的预算控制策略。显式选择真实模型时，插件不干预路由，请求原样通过。 该插件要求 Node 版本 ^22.19.
Read Moredsh-legion：DeepSeek Harness 的多智能体委派与模型路由插件
dsh-legion 是一个专为 DeepSeek Harness (DSH) 设计的 TypeScript 多智能体编排与 LLM 模型路由插件，旨在扩展现有运行时能力，而非引入独立框架。其核心机制是将委派策略交由部署侧配置，协调代理仅通过 `legion` 工具选择语义化的 Specialist（如 `quick`、`deep`、`review`），具体的模型路由、后端类型、工具过滤、深度及结果契约均由插件配置决定。该插件支持精确的多提供商模型路由、多种子代理后端策略、结构化版本化结果输出、自定义 Cohorts 以及声明式 Strategies，并具备硬资源限制和可解释性诊断功能。安装过
Read Moredsh-opencode-go-provider：DSH rc.6 的 OpenCode Go provider 插件
`dsh-opencode-go-provider` 是 `edynasty` 维护的 MIT 许可插件，专为 DeepSeek Harness（DSH）`rc.6` 设计。它提供 `opencode-go` 模型路由、Web Connect 设置卡片、stale-while-revalidate 模型目录及安全诊断命令。该插件未发布至 npm，需通过固定 commit 的 Git 依赖方式安装到 DSH `web` profile。运行环境要求 Node.js `^22.19.0 || >=24.0.0`、pnpm `11.7.0` 及有效的 OpenCode Go API key，后者仅通
Read Moredsh-switch: A Model Control Plane Plugin for DeepSeek Harness
dsh-switch is an evidence-first model control plane plugin for DeepSeek Harness (DSH). It provides two independent functions: first, within DSH, it checks registered Provider routes and the default model via `dsh_switch_status`, and verifies and saves the default model settings for new sessions via `dsh_switch_default`, affecting only new sessions and never intercepting or rerouting active session requests; second, it offers a standalone CLI tool for read-only probing of OpenAI-compatible `/models` endpoints, reporting reachability, HTTP status, latency, model counts, and credential configuration status, without consuming inference tokens or verifying end-to-end generation capability. The plugin emphasizes security boundaries and rejects embedding...
Read Moredsh-model-router: A Thin Routing Strategy Plugin for DeepSeek Harness (DSH)
dsh-model-router is a thin routing policy plugin for DeepSeek Harness (DSH), version v0.2.2, licensed under the MIT License. It adds rule execution on top of DSH's existing model management, with core features including deterministic keyword routing, the whitelist-based model switching tool `model_route` that is disabled by default, and an isolated Beta image generation channel `image_gen`. Routing rules are executed in configuration order, stopping at the first match, while preserving the current state if no rule matches; model switching requires precise validation of provider, model, and other fields. `image_gen` uses an OpenAI-compatible endpoint, enforces HTTPS and strictly restricts result downloads, and supports integration with third-party services such as Seedream; however, video, audio, and 3D features are not included.
Read Moredsh-model-router: Model Routing and Cost Dashboard for DeepSeek Harness
`dsh-model-router` is a model routing and cost optimization plugin for DeepSeek Harness. It routes simple requests to cheaper models through heuristic rules, reserves complex requests for the main model, and automatically falls back on instantaneous failures. This approach reduces costs while ensuring quality for requests requiring context, vision, or reasoning capabilities. The plugin supports vision-aware routing, prioritizing available vision models if messages contain images, and respects user-selected models. In automatic mode, it prompts the user for a quick response when a simple request is detected. Additionally, it provides a real-time panel in the Composer Dock, displaying token usage, cache hits, and estimated costs based on the pricing table. Installation requires ensuring the `dsh` CLI is in the PATH, followed by restarting the Web profile.
Read Moredsh-tier-router: Hierarchical Model Routing for DeepSeek Harness
`dsh-tier-router` is a hierarchical model routing plugin for DeepSeek Harness (DSH), designed to optimize cost and efficiency through "premium-tier" and "economy-tier" strategies. The premium-tier models handle complex tasks such as planning, architecture, and review, while the economy-tier models manage routine implementations. Core features include automatic hierarchical routing (defaulting to `auto` mode), session-level scope switching, on-demand consultant and review tools, automatic escalation on failure, ordered fallback chains, dynamic adjustment of reasoning intensity, gating for high-impact operations, and hierarchical delegation to sub-agents. The plugin supports flexible configuration of models, effort levels, and sub-agent strategies via `/tier` series commands, with both persistent and temporary configuration capabilities. Released under the MIT license, installation requires meeting specific Node.js version requirements and syncing to the DSH discovery root directory. It is suitable for applications needing
Read More《Cursor Docs》-Cursor Router
Cursor Router is the model routing system behind Auto . Not every request needs frontier level intelligence, so the router sends each request to the model that fits the task: simple requests go to fast, efficient models while complex work goes to the most capable ones.
Read MoreUse dsh-model-router to implement model routing and cost dashboard for DeepSeek Harness
`dsh-model-router` is a DeepSeek Harness web plugin maintained by *tianji-qingtian*. It directly answers simple questions after verification by the Flash Judge, automatically downgrades when instantaneous failures occur, and displays token counts, cache hit status and estimated costs below the input box. This article verifies the installation commands, `/router` usage and known limitations by cross-checking with the directory page and GitHub README/source code.
Read MoreOmniRoute Sees Over 9,000 Star Gains in a Week: Unifies 290+ Model Providers via Local AI Gateway
OmniRoute (diegosouzapw/OmniRoute) is one of the open source projects with the fastest growth in stars in the GitHub AI Agent category in July 2026, with an increase of approximately +9,200 in 7 days. This MIT-licensed local AI gateway aggregates over 290 model providers (including over 90 free tiers) through a single OpenAI-compatible endpoint, with built-in Combo automatic fallback, quota-aware routing, cost telemetry, and support for MCP/A2A. It can connect to programming agents such as Claude Code, Cursor and Codex. This article introduces the pain points of multi-model switching it solves, the core routing mechanism, quick start steps, as well as the security and trust boundaries during deployment.
Read More