Foreword

In DSH’s plugin system, model reasoning capabilities can be supplemented through plugins. For third-party models in llm-pi-ai, if a model entry does not declare reasoningEfforts, default effort levels need to be added, and DSH effort levels should be mapped to gateway values, for example, high maps to ultra.

hytime/dsh-thinking-effort is a DSH plugin that provides a configuration layer for this purpose: it supplements configurable thinking effort levels for these models and sets the default thinking effort for sub-agents.

What This Is

The repository path is hytime/dsh-thinking-effort, the npm package name is @hytime/dsh-thinking-effort, the repository owner is hytime, and the license is MIT. It is aimed at DSH users who need to manually configure reasoning effort for llm-pi-ai models.

Core Features

  • Adds default off, high, and max effort levels for models that do not declare reasoningEfforts, without overwriting custom values.
  • Allows selecting reasoning effort levels per model in DSH Settings and configuring corresponding gateway values.
  • Maps DSH effort levels to gateway values, for example, high maps to ultra.
  • Sets the default thinking effort for sub-agent requests, which only applies if the request does not explicitly specify one.
  • Includes Chinese, English, Japanese, and Korean setting dictionaries; Japanese/Korean switching requires DSH core locale support.
  • Displays a watermark of the installed plugin version in the bottom-right corner of the settings page.

Installation and Enabling

Use the official DSH CLI to manage plugin profiles. A standard npm install will not register the DSH profile bundle.

dsh plugin --profile <profile> add @hytime/dsh-thinking-effort

The above command adds the plugin to the specified DSH profile. Host changes require restarting DSH; settings and locale changes take effect in the browser, and you may need to refresh the page if necessary.

Typical Usage

  1. Open Settings → Model capabilities and effort in DSH.
  2. On Stock DSH, you can use Page language to select 中文 or English; 日本語 and 한국어 require DSH core locale support.
  3. In the Subagent default effort card, select the default effort level for sub-agents and click Apply.
  4. Use Quick settings to apply the official DeepSeek or general presets for all models, or expand provider/model for detailed configuration.
  5. Select the reasoning effort level and fill in the precise gateway value.
DSH Effort Level Gateway Value
off Leave blank to omit the parameter
high ultra
max max
  1. Return to Composer, select a model, and use its reasoning effort selector.

Applicable Scenarios and Notes

This plugin is suitable for the following situations:

  • Using llm-pi-ai to connect third-party models;
  • The model does not declare reasoningEfforts, and reasoning effort levels need to be supplemented per model;
  • Mapping DSH effort levels to gateway values;
  • Setting a default thinking effort for sub-agents while preserving explicitly requested values.

Notes before use:

  • llm-pi-ai exposes seven standard effort levels: off, minimal, low, medium, high, xhigh, and max.
  • Non-off levels require a gateway value; an empty off value indicates omitting the parameter.
  • The selected sub-agent effort level must be supported by the target model, otherwise the gateway may return UNSUPPORTED_REASONING_EFFORT.
  • Both off and unset effort levels may omit the reasoning field; whether thinking is disabled depends on the gateway protocol.
  • Version 0.1.8 includes Japanese and Korean dictionaries, but the current official DSH only exposes zh and en via LocaleRuntime; selecting ja or ko natively in DSH will fail.
  • Modifying only this plugin cannot extend DSH’s global locale list.
  • The plugin runs with the permissions of the current dsh process; review the source code and license before installation.

Reference Links

GitHub repository:

https://github.com/hytime/dsh-thinking-effort

Directory page clue (not confirmed in this material):

https://www.skillhub.cn/plugins/hytime/dsh-thinking-effort