Preface

When integrating multiple model providers in DeepSeek Harness (DSH), the context window, maximum output, input modalities, and reasoning tiers for a single model are often scattered across different configuration sections. Manually consulting documentation and filling in each field is slow and can lead to inconsistencies with the compatibility fields required for the openai-completions inference dispatch. To reuse parameters from the llm-pi-ai preset or other configured providers, one typically has to cross-reference and transcribe settings from multiple places.

The dsh-provider-model-configurator, maintained by community contributor LiangYin233, consolidates the above capabilities into a single interface on a dedicated settings page titled “Model Pro” within the DSH WebUI. It allows you to view, create, edit, copy, and delete model entries under each provider, and supports one-click form population from the pi-ai preset or any other configured provider.

What Is This

dsh-provider-model-configurator is a web-side plugin for DSH, classified as model inference. It provides a graphical model entry management interface under the “Model Pro” navigation item on the left side of the settings page (located after the Models page), covering the context window, maximum output, input modalities, reasoning tiers, and the inference compatibility switch (compat).

The current plugin version is v0.3.9, released under the MIT license, and is available on both the SkillHub Community Directory and GitHub. SkillHub is an independent community site with no official affiliation to DeepSeek / High-Flyer; the DSH ecosystem follows an “everything is a plugin” philosophy, and such tools are maintained by community contributors.

Core Features

CRUD Operations for Model Entries

After selecting a target provider, the plugin lists all explicitly configured model entries and configuration summaries under that provider. You can edit or delete individual entries.

When creating a new model, enter the model ID, then fill in the display name, context window, maximum output, input modality (text / image), and reasoning tier (the tier corresponds to the wire value; off means leave blank and not sent). Both reasoning tiers and input modalities support “Unset (inherit from directory)” to avoid writing unconfigured fields with explicit default values.

Copy & Populate from Presets or Other Providers

Click “Use Model Preset” to open a source selector. You can choose a model from the built-in llm-pi-ai model list or any other configured provider to quickly populate the form. The built-in directory reads the target model’s context parameters and maximum tokens, eliminating the need to manually consult documentation and fill in these fields.

Inference Compatibility Switch (compat)

Configure thinkingFormat (valid values include openai, deepseek, openrouter, together, zai, qwen, string-thinking, ant-ling) and supportsReasoningEffort (true / false / Unset) for the openai-completions inference dispatch to read.

Directory Snapshot Behavior

When creating or applying a model on a directory provider that does not yet have an explicitly configured model list, the plugin writes the current model list from the built-in directory as the explicit entries for that provider: existing IDs in the directory are updated in place, and new IDs are appended. Subsequent edits are based on this explicit list. Later updates to the directory will not automatically merge into the snapshotted list. If all explicit entries for a directory provider are deleted, it reverts to the built-in directory (unsnapshotted) state. Be mindful of this behavior when using directory-type providers to avoid mistakenly thinking that directory updates will automatically sync to the snapshotted list.

Installation and Enablement

The official README provides the following installation methods (using v0.3.9 as an example):

# Install from GitHub
dsh plugin --profile web add github:LiangYin233/dsh-provider-model-configurator#v0.3.9

# Or install from a GitHub tarball
dsh plugin --profile web add https://github.com/LiangYin233/dsh-provider-model-configurator/archive/refs/tags/v0.3.9.tar.gz

# Or install from a local pack
npm pack
dsh plugin --profile web add ./dsh-provider-model-configurator-0.3.9.tgz

After installation, restart the web server and refresh the page. Open Settings and navigate to “Model Pro” on the left sidebar.

Typical Usage

  1. Install the plugin and restart the DSH web service. Refresh the browser and go to Settings → “Model Pro”.
  2. In the interface, select the target provider and view the existing list of model entries and their configuration summaries.
  3. Create a new model: enter the model ID and manually fill in each field; or click “Use Model Preset”, select a source from the pi-ai preset directory or another provider, and populate fields like context window and output limit with one click.
  4. Set the reasoning tier and input modalities as needed; leave uncertain fields as “Unset (inherit from directory)”.
  5. If the provider uses openai-completions for inference dispatch, configure thinkingFormat and supportsReasoningEffort in the compat section.
  6. After saving, the explicit model list for the target provider is updated. For directory providers, the first operation may trigger a directory snapshot, and all subsequent edits are made to the explicit list after the snapshot.

Use Cases and Notes

Who It’s For

  • Developers managing multiple providers and models in the DSH WebUI who need to maintain context windows, output limits, and reasoning-related fields in a unified manner.
  • Users who want to quickly apply parameters from the llm-pi-ai preset or copy model configurations from existing providers to reduce redundant data entry.
  • Scenarios requiring explicit configuration of thinkingFormat and supportsReasoningEffort for the openai-completions dispatch path.

Before Using

  • The plugin runs with the permissions of the current DSH process. Before installation, review the source code and the MIT license to ensure it aligns with your deployment policies.
  • Directory providers will form a snapshot after the first write; subsequent updates to the built-in directory will not auto-merge. If you rely on auto-updates from the directory, understand how to revert and reconfigure.
  • The version tags in the installation commands (e.g., v0.3.9) are subject to the README and release page. When upgrading, replace the corresponding tag or tarball URL.

Conclusion

dsh-provider-model-configurator consolidates the scattered model parameters and inference compatibility configurations in DSH into the “Model Pro” settings page and supports one-click population from the pi-ai preset or other providers. It is well-suited for scenarios requiring frequent adjustments to model entries across multiple providers. For more information, see the SkillHub Directory Page and the GitHub Repository.