Introduction¶

When handling models for custom providers in DSH, some inference levels need to be written into the llm-pi-ai configuration. If you only want to adjust the reasoning effort of a single model in the Web UI instead of manually editing the configuration file every time, you can make this kind of settings page into a plugin.

DSH Effort Tweak is a small DSH plugin provided for this scenario. The following introduces its purpose, installation method, typical usage, and things to note.

What is it¶

toukaiteio/dsh-effort-tweak is a DSH plugin maintained by Toukaiteio with an MIT license.

It provides a settings page for editing reasoning effort by model for DeepSeek Harness custom providers, allowing you to view configured models in the DSH Web UI and select reasoning levels for a single model without directly editing the configuration text.

Key Features¶

• Adds a Settings → Reasoning effort page in the DSH Web UI.
• Lists models explicitly configured in llm-pi-ai.providers.<provider>.models.
• Supports selecting off, minimal, low, medium, high, xhigh, or max for a single model.
• Preserves other fields of the model, as well as existing provider-specific wire values.
• Saved via DSH’s settings service, eliminating the need to manually edit configuration files.

Installation and Enablement¶

The installation script will download the latest GitHub Release archive, verify the SHA-256 provided by GitHub when available, and keep the archive under DSH_HOME/plugin-archives/dsh-effort-tweak/ for future dependency resolution. The script will install the plugin via dsh plugin. Unless --no-start or -NoStart is passed, the script will start the default web Profile.

Windows PowerShell¶

Execute in PowerShell:

irm https://raw.githubusercontent.com/Toukaiteio/dsh-effort-tweak/main/scripts/Install-DshEffortTweak.ps1 | iex

macOS and Linux¶

Execute in terminal:

curl -fsSL https://raw.githubusercontent.com/Toukaiteio/dsh-effort-tweak/main/scripts/install-dsh-effort-tweak.sh | bash

Local Package Installation¶

You can also install from a local package. The command below is an example of local package installation; it will install dependencies, build, and then add the local .tgz package to the web Profile:

pnpm install
pnpm build
dsh plugin --profile web add ./dsh-effort-tweak-0.1.0.tgz

After installing or updating the plugin, restart the DSH Web to load the new client bundle.

Typical Usage¶

First add a model list for the custom provider, then open Settings → Reasoning effort. This page does not display catalog-only providers, so if the target provider or model is not visible on the page, usually first check whether there is already a model list under that custom provider.

The configuration will be reflected in llm-pi-ai’s providers.<provider>.models. The example shape is as follows:

providers:
  my-gateway:
    models:
      - id: my-reasoning-model
        reasoningEfforts:
          off: null
          low: low
          high: high

In this example, reasoningEfforts is used to configure the available effort mappings for that model. After selecting a level via the page, saving is completed through DSH’s settings service.

Applicable Scenarios and Notes¶

Suitable for the following scenarios:

• Models lists have already been configured in DSH custom providers.
• Wish to adjust reasoning effort individually per model.
• Do not want to manually modify llm-pi-ai.providers.<provider>.models related fields in the configuration file every time.

Notes:

• The page does not display catalog-only providers. You need to first add a model list in the custom provider before opening the page to adjust models.
• The plugin runs with the permissions of the current dsh process. You should check the source code and MIT license before installing.
• Local package installation commands will reference the local .tgz package; the actual file name depends on your local build artifacts.
• The development environment requires Node.js 22.19 or later and pnpm 10.

Conclusion¶

The function of DSH Effort Tweak is relatively focused: placing the selection of reasoning effort for custom provider models into the DSH Web UI, and saving it through the DSH settings service, reducing manual configuration editing operations.

GitHub repository:

https://github.com/Toukaiteio/dsh-effort-tweak

The community directory is an independent site with no official affiliation to DeepSeek / Huansuan, nor is it an official app store. This article only uses verified information and does not provide URLs for unverified directory pages; if needed, you can search for the plugin name dsh-effort-tweak in the community directory.