DSH 模型补齐助手：新渠道保存后自动拉满上下文与多模态
介绍社区插件 dsh-model-auto-enricher（66rapper）：在 DSH 添加模型渠道并保存后，自动补齐上下文窗口、最大输出、文本+图片多模态与 low/high/max 思考档位，支持系列识别与未知模型兜底规则，写前时间戳备份且只升不降。含插件市场与 desktop/web 官方安装命令，以及「模型补齐助手」管理面板用法与适用注意。
Read Moredsh-model-capability 插件介绍
`dsh-model-capability` 是一款由 iuuuuuuuu 维护的 DeepSeek Harness 插件，旨在解决原生手动编辑配置繁琐且易丢失的问题。该插件在 DSH 的「设置 → 模型」页面提供直观的界面，支持逐模型配置视觉能力、推理档位及上下文容量。 核心功能包括：自动格式化容量数值（支持 K/M 写法）、直接写入官方配置无需重启，以及自动备份最近 12 份配置以防止数组写入时注释丢失。安装命令为 `dsh plugin --profile desktop add "git+https://github.com/iuuuuuuuu/dsh-model-capability
Read MoreDeepSeek Harness Plugin: dsh-subagent-model-config
This plugin aims to resolve the issue that the native Team formation in DeepSeek Harness (DSH) cannot assign models to individual team members separately. It wraps internal DSH interfaces to provide a GUI, enabling users to configure models, inference levels, and security rules for specific members, with hot-reload support. The plugin includes features such as Glob matching, a dropdown for inference-level selection, forensic logging, and capability restrictions during video presets, all of which take effect without restarting DSH.
Read MoreDeepSeek Harness Plugin: dsh-models-input-modalities
The `dsh-models-input-modalities` plugin addresses configuration pain points by exposing previously hidden model capability fields in the DeepSeek Harness Web interface. It allows users to directly select input modalities (text/images) and reasoning levels without manually editing YAML. It supports localization, conflict handling, and configuration migration, providing users with more intuitive configuration control.
Read Moredsh-model-autoconfig: Automatic Correction of Model Thinking Tiers and Capabilities
The DSH model catalog often exhibits abnormal thinking mode due to missing or incorrect configurations. Maintained by wangser10032, the dsh-model-autoconfig plugin is designed to automatically complete these configurations. It supports the official API and multiple third-party platforms, and can automatically correct thinking levels, set configuration parameters, complete model capabilities, and check for errors. After installation, restart dsh web and add a model in Settings to automatically complete the configuration. It also supports remembering users' selected thinking levels, making it suitable for developers who frequently configure models.
Read MoreIntroduction to SupaNexus Core Plugins
`@supanexus/dsh-plugin-supanexus-core` is the core plugin of the SupaNexus platform, designed to unify brand styling, streamline model configuration, and manage the application ecosystem. Its main features include brand UI customization, quick OAuth authorization, dual-channel model access, and a sidebar application plugin marketplace. Open-sourced under the MIT license, it requires a service restart after installation to take effect. When using the plugin, pay attention to data retention and permission security, and note that it has no official affiliation with DeepSeek.
Read MoreDeepSeek Harness Plugin: dsh-plugin-model-config
`dsh-plugin-model-config` is a plugin designed for the DeepSeek Harness (DSH) Web interface to address the lack of detailed Adapter configuration options on the default settings page. It displays configuration items such as inference mode, context window, and budget limits as rows within the Provider card, supporting control types including number, toggle, text, and dropdown. The plugin dynamically generates the interface by reading the deployment Schema, allowing new configurations to be accommodated without modifying source code. Users must install it with a package manager and restart the process for changes to take effect. Before installation, the code should be reviewed; note that the plugin supports the Web interface only, and direct modifications to in-memory configurations cannot be undone.
Read Moredsh-awesome-model-setting: DSH Model Capabilities Graphical Configuration Plugin
The DSH plugin `dsh-awesome-model-setting` transforms model capability configuration from tedious manual editing of `settings.yaml` into a graphical workflow. Users can visually manage input modalities, context window size, maximum output tokens, and image quotas in the settings panel. The plugin supports loading model lists from providers and displays the merged effective values of configurations and adapter catalogs in real time. It also provides “Restore Route Defaults” and “Reset to Default Parameters” options for easy rollback. Installation requires the official command and a service restart. It is suitable for scenarios requiring fine-grained management of multimodal model configurations, simplifying the process through a user interface.
Read MoreDeepSeek Harness Plugin: dsh-model-modality
The DSH plugin `dsh-model-modality` addresses the issue of third-party models being unable to automatically detect image input in DeepSeek Harness. It allows users to explicitly declare model metadata (`inputModalities`) to ensure that images, attachments, and other features work correctly. The plugin supports writing configurations for both official and third-party providers, provides declaration and query tools, and offers an optional browser-side UI. The installation command is `dsh plugin add ct-jyjntc/dsh-model-modality`, and it requires a specific version of `@deepseek-ai/dsh-tools`.
Read Moredsh-model-info-hint: Display Model Configuration Details on Hover
The `dsh-model-info-hint` plugin is designed to improve the efficiency of viewing model configurations in DeepSeek Harness (DSH). It reads the `~/.dsh/settings.yaml` configuration file and, when hovering over the model selector, non-blockingly displays detailed model information such as context window size, output tokens, and authentication method. The plugin consists of a Host-side component and a browser-side component and supports hot reload. To install it, add the directory as a local package to the web profile; after restarting, configurations can be viewed by hovering over the model selector.
Read Moredsh-subagent-model: Independent Model and Preset Configuration for DSH Subagents
The DeepSeek Harness plugin is designed to address the resource overhead caused by a main agent invoking subagents that inherit the main model. It adds a "Subagent Model" menu, allowing users to uniformly configure independent providers, models, and reasoning effort for all delegated subagents. The plugin provides official UI integration, a globally applied delegation policy, and preset tier management, while supporting foreground waiting and context inheritance. Installation requires mounting an npm package and configuring dependencies; configuration is stored in a local JSON file and is designed for the DSH Web environment.
Read Moredsh-subagent-setting: Unified configuration of subagent default model in web settings page
## Introduction If you run multi-subagent task flows in DeepSeek Harness (DSH), you might encounter a specific issue: the default configurations for a subagent's provider, model, and reasoning effort are scattered across configuration files. To switch to a different default model, you must manually edit the YAML file and then verify that every subagent has adopted the new values. `dsh-subagent-setting` brings this to a web settings page: select the default provider, model, and reasoning effort on the interface, and newly created subagents will take effect immediately. You can also choose to apply this to existing subagents.
Read Moredsh-models-dev-reasoning: Fill missing thinking levels for unconfigured DSH models
dsh-models-dev-reasoning is a DeepSeek Harness (DSH) plugin maintained by aerince, licensed under the MIT license. This plugin aims to address the issue where third-party or custom models in llm-pi-ai lack the reasoningEfforts configuration, resulting in the inability to display thinking levels. By fetching data from models.dev, it automatically completes missing reasoning levels for models that have not yet set this parameter, ensuring the model selector displays correctly without overwriting existing configurations. The plugin workflow is as follows: after waiting for settings to be available, it prioritizes reading models.dev/api.json, falling back to a GitHub tarball if that fails; subsequently, it matches models under llm-pi-ai and fills in missing reasoning levels, while also copying the directory...
Read MoreSeverusZh/dsh-plugin-subagent-director: Assigning Models and Role Division for DSH Sub-agents
SeverusZh/dsh-plugin-subagent-director is an MIT-licensed plugin for DeepSeek Harness designed to address issues such as redundant LLM configurations for sub-agents and lack of planning in role allocation. Its core functionalities include specifying LLM providers and models for different sub-agents, as well as defining sub-agent responsibilities, personas, and optional model bindings through "role templates," thereby assisting the main agent in delegating tasks effectively. The plugin supports a four-level fallback chain (single call > role binding > plugin default > inheriting from the main agent) and provides hot-reloading for configurations. Users can define roles such as "translator" via `settings.yaml`, utilize the `subagent_role` tool for delegation, and explicitly specify models to override defaults during single calls.
Read MoreDeepSeek Harness Model Configuration Plugin: Managing Custom Endpoints and Model Parameters in Web UI
marvekg/deepseek-harness-model-config adds a "Model Advanced Configuration" page to the DeepSeek Harness Web UI, supporting the creation of OpenAI/Anthropic protocol endpoints, pulling or manually adding models, and populating context and inference parameters via models.dev. API keys are stored through Harness credentials and are not written to settings.yaml. Installation: dsh plugin --profile web add github:MarvekG/deepseek-harness-model-config, usable under "Settings → Model Advanced Configuration." Licensed under MIT, maintained by MarvekG.
Read MoreAdd "Model Pro" to DSH Web UI with dsh-provider-model-configurator
The DSH interface enhancement plugin `dsh-provider-model-configurator`, maintained by LiangYin233, adds a "Model Pro" page in the Web UI settings. It allows users to centrally view, create, edit and delete model entries of the llm-pi-ai provider, and can copy context window, maximum output and inference gear parameters from pi-ai presets. This article verifies the installation command, directory snapshot behavior and precautions based on the community directory page and the GitHub v0.3.9 source code.
Read More