Preamble

The model catalog bundled with DSH is not complete; thinking tiers and capability settings for some models may be missing or incorrect. Taking the pi-ai model catalog as an example, the absence of the off tier can cause dsh to misjudge a model’s thinking state, potentially preventing thinking mode from being enabled or disabled correctly, and even causing long outputs to be unexpectedly truncated. The dsh-model-autoconfig plugin aims to automatically fill in these missing configurations.

What Is This

This is a plugin that automatically corrects dsh model thinking modes and model capabilities, maintained by wangser10032. It supports official APIs as well as third-party platforms or proxies such as Coding Plan, SiliconFlow, Volcano Ark, Alibaba Bailian, and OpenRouter.

Core Features

  1. Correct Thinking Mode
    - Checks whether a model supports thinking and whether thinking can be disabled.
    - Fills in supported tiers (e.g., Low, High, Max).
    - Remembers the last selected thinking tier for each model.

  2. Automatically Configure Model Parameters
    - Populates correct parameters based on the model vendor’s official documentation.

  3. Complete Model Capabilities
    - Fills in capabilities such as vision, context length, and maximum output length.

  4. Check Incorrect Configurations
    - Identifies invalid or misleading thinking tiers.
    - Detects overlooked vision capabilities or configurations that would be rejected by dsh.

Installation and Activation

Run the following command in the terminal to install it:

dsh plugin --profile web add dsh-model-autoconfig

After installation, you must restart dsh web for the plugin to take effect.

Usage

  1. Open dsh’s Settings → Models.
  2. Add a model service, and fill in the API address and API Key.
  3. Retrieve the model list or manually enter the model ID (if the list API is unavailable), then save.
  4. The plugin will automatically complete the remaining configuration; no manual editing of settings.yaml is required.
  5. After selecting a thinking tier for a model, the plugin will remember the setting in the current browser. When you switch back to that model, the last tier will be restored automatically.

Dependencies

This plugin depends on the following peerDependencies:
- @deepseek-ai/schemastery
- @deepseek-ai/dsh-credentials
- @deepseek-ai/dsh-client-ui-model-selection

If the changes are not reflected immediately, refresh the webpage or restart dsh web.

Conclusion

By automatically correcting configurations, this plugin resolves functional issues caused by incomplete dsh model catalogs, making it suitable for developers who frequently configure different model capabilities.

Plugin catalog: https://www.skillhub.cn/plugins/wangser10032/dsh-model-autoconfig

Code repository: https://github.com/wangser10032/dsh-model-autoconfig