Preface

DeepSeek Harness (DSH) intentionally hides some per-model configuration items in its default settings interface, including the Efforts setting for reasoning mode. The lylarcher/dsh-model-capabilities plugin directly addresses this issue. It provides a settings page that allows you to directly edit each model’s input modality and reasoning Efforts without manually editing the settings.yaml file. This plugin is available on both DSH Web and DSH Desktop.

Core Features

The plugin adds a “Model Input and Reasoning” settings area, with the following main capabilities:

  • List all configurations: Enumerates all configured providers, their models[] entries, and modelOverrides, providing a full configuration overview.
  • Toggle input modality: Switches the input modality for each model (for example, text or image).
  • Set reasoning Efforts: Configures reasoningEfforts for each model. You can choose to disable reasoning mode (writes false) or configure wire values for levels such as minimal, low, medium, high, xhigh, and max. At least one non-off level must be provided.
  • Write configuration directly: Writes directly to the llm-pi-ai settings namespace; the runtime adapter re-registers routes so the model selector immediately shows the new reasoning levels.

Installation and Activation

Use the official installation command to add the plugin to the specified profile:

dsh plugin --profile <name> add dsh-model-capabilities

After installation, restart DSH Web or DSH Desktop to load the new plugin.

Usage and Limitations

  • Prerequisite: If the plugin page is empty, first configure the provider and model in the built-in Models page.
  • Plugin choice: If you only need to choose the input type (text/image) when adding a model, the lightweight WJZ-P plugin is the better option. If you need to configure reasoning Efforts and view all configurations (including overrides) at the same time, use this plugin.
  • Permissions and conflicts: The plugin runs with the permissions of the current DSH process. In a read-only connection, the page displays a notification and disables saving; if a write conflict (stale revision) occurs, it prompts you to refresh the page and retry.
  • Design limitation: Not exposing these fields is an intentional design choice in the built-in Models page, and this plugin fills that gap.

Summary

The dsh-model-capabilities plugin directly manages each model’s input modality and reasoning Efforts through the UI, filling the gap for configuration items missing from DSH’s default settings. If you need fine-grained control over reasoning mode, this plugin provides the necessary tool.

For more details, refer to:
* GitHub repository
* Plugin directory