Introduction

The WebUI plugin architecture of DeepSeek Harness (DSH) allows developers to extend the capabilities of the native settings page. In model inference scenarios, developers often need to manage configurations for both the official DeepSeek provider and custom LLM providers, while fine-tuning inference parameters. This plugin aims to fill this gap by adding an “Advanced Model Settings” entry to the DSH WebUI.

Plugin Overview

This is a DSH WebUI plugin maintained by u9521, primarily used for managing LLM provider configurations on the WebUI settings page.

It addresses the need to uniformly manage model parameters across different namespaces in a single interface, including the custom namespace llm-pi-ai and the official namespace llm-deepseek. The plugin supports built-in and custom providers, and covers configuration items such as model lists, reasoning/thinking options, request headers, and retry policies.

Core Features

The core capabilities of the plugin focus on model configuration management:

  • Multi-namespace management: Supports managing configuration profiles in the custom provider namespace (llm-pi-ai) and the official provider namespace (llm-deepseek).
  • Providers and models: Supports both built-in providers and custom providers, with model list management.
  • Inference parameters: Provides configuration options related to reasoning/thinking.
  • Requests and fault tolerance: Supports custom request headers and retry policies.
  • Rule consistency: Follows the validation rules and conflict-handling logic of the official DSH settings UI.

Installation and Activation

After installing the plugin, you need to restart the Web service and refresh the browser cache.

  1. Run the installation command to pull the pre-built version from the dist branch on GitHub:
    dsh plugin --profile web add github:u9521/dsh-advanced-model-editor#dist

  1. After installation, restart the Web service (dsh web) and perform a hard refresh in your browser (Cmd+Shift+R / Ctrl+Shift+R).

  2. In the WebUI settings page, look for the entry named Advanced Model Settings.

Upgrading

When a new version is released, you can pull the latest dist release using the update command.

  1. Run the following command to update remotely:
    dsh plugin --profile web update @local/dsh-advanced-model-editor

Alternatively, if re-adding the dist branch:

    dsh plugin --profile web add github:u9521/dsh-advanced-model-editor#dist
  1. You also need to restart the Web service and perform a hard refresh in the browser.

Environment Requirements and Notes

Before installing or using this plugin, make sure your environment meets the following dependencies:

  • DSH version: DSH version >= 0.1.5-rc.1 is required. The main branch of the plugin is no longer backward compatible with DSH versions earlier than 0.1.5.
  • Node.js: Node.js version ^22.19 || >=24 is required.
  • Package manager: pnpm is required, because the dsh plugin command internally depends on pnpm for building and adding operations.

Conclusion

By providing a structured entry for model parameter configuration in the DSH WebUI, this plugin simplifies the management of custom LLM providers and inference parameters. All configurations are based on the official DSH validation rules, ensuring configuration stability.

  • Plugin directory: https://www.skillhub.cn/plugins/u9521/dsh-advanced-model-editor
  • Source repository: https://github.com/u9521/dsh-advanced-model-editor
  • License: MIT