Introduction

In the DeepSeek Harness (DSH) plugin ecosystem, configuring model capabilities typically involves declaring provider routes. For DSH users utilizing OpenAI-compatible providers, fields such as reasoningEfforts and input must be declared in the settings page; otherwise, the model selector cannot display effort controls, and image conversations might be rejected if the model has not declared image.

better-model-provider addresses this issue by turning per-model capability declarations into a settings section, reducing the need for direct YAML editing. It is maintained by sanshanya and licensed under MIT. The DSH community directory is an independent site, with no official affiliation to DeepSeek or High-Flyer.

What This Is

better-model-provider is a DSH web plugin designed to declare and edit capabilities for individual models within DSH settings. It allows editing declarations on custom models and saving sparse overrides on official catalog routes; provider configurations remain managed on the official Models page.

Verified compatibility is declared for the dsh 0.1.x line, with the contract:

@deepseek-ai/dsh-api-remotes >=0.1.0-rc.7 <0.2.0

Core Features

Below are the verified main capabilities.

Editing Reasoning Effort

  • Declare reasoningEfforts for a model.
  • After selecting Custom, you can check the required levels.
  • If “off” is needed, check the off option.

Editing Input Modalities

  • Declare request modalities, such as text / image.
  • Vision models need image checked under input modalities; otherwise, sending images will be rejected.

Editing Token Capacities

  • Edit contextWindow / maxTokens.
  • Supports K/M notation, e.g., 380K, 1M.
  • Changes take effect immediately after clicking Apply.

Official Catalog Routes

  • Edit official models via Manage official models.
  • Only saves differences from official defaults; other parts continue to follow catalog updates.
  • Use Reset to official defaults to revert all edits for a single model.

Unconfigured Routes

  • Expand installed-but-unconfigured routes via Manage official providers (N).
  • Select an item and apply the first modification to create a route.
  • API keys still need to be configured on the official page.

Boundaries

  • Dedicated-adapter apps (built-in DeepSeek / OpenAI Codex) declare capabilities in their own settings pages and do not appear on this plugin page.
  • Surfaces outside the contract degrade silently.

Installation and Enablement

The plugin runs with the current dsh process permissions. Before installation, please review the source code, license, and injected scripts. After confirming use, execute:

dsh plugin --profile web add github:sanshanya/better-model-provider#master

After installation, restart the web:

Restart dsh web

The Settings sidebar will have a new Model capabilities section.

If installing the default branch, it will build from source and print a pnpm allowBuilds key that needs to be added, then re-run add. For local link: installs, first run:

npm install && npm run build

Uninstall:

dsh plugin --profile web rm better-model-provider

Typical Usage

Below is the operational sequence.

  1. First, configure providers and API keys on the official Models page. This plugin page does not manage keys or route lifecycle again.
  2. Expand model rows and edit capabilities:
    - For Reasoning effort, select Custom and check the required levels; check off if needed.
    - For vision models, check image under input modalities.
    - In Capacities, you can enter 380K, 1M, then click Apply.
  3. Official models are edited via Manage official models. Only differences are saved, and you can revert using Reset to official defaults.
  4. For installed-but-unconfigured routes, expand via Manage official providers (N), select an item, and apply the first modification to create the route; API keys are still configured on the official page.

Use Cases and Notes

Suitable for users managing OpenAI-compatible provider model capabilities in DSH web, especially when needing to adjust reasoning effort, image input, and token capacities per model.

Notes:

  • API keys, providers, and route lifecycle are still managed on the official Models page.
  • This plugin page only edits capability declarations and official route differences.
  • Surfaces outside the contract degrade silently.
  • The plugin runs with permissions equivalent to the current dsh process; review the source code and license yourself before installation.

Links

GitHub: https://github.com/sanshanya/better-model-provider

Directory page: https://www.skillhub.cn/plugins/sanshanya/better-model-provider