Introduction

In DeepSeek Harness (DSH), a common scenario when using third-party or custom models is that the model is available in llm-pi-ai but lacks reasoningEfforts, so the model selector cannot see the reasoning tiers.

The dsh-models-dev-reasoning plugin handles this: it fills in models that haven’t set reasoning tiers from models.dev without overwriting existing configurations.

What is this

dsh-models-dev-reasoning is maintained by aerince and is licensed under MIT.

It adds models.dev reasoning tiers to third-party or custom models in DeepSeek Harness that have not yet set reasoningEfforts, allowing the model selector in llm-pi-ai to display reasoning tiers.

Core Features

  • Fills in reasoning tiers for third-party or custom models in DSH from models.dev.
  • Enables the model selector in llm-pi-ai to display reasoningEfforts.
  • Only processes models that have not yet set reasoningEfforts; existing configurations will not be overwritten.
  • Maps none from models.dev to off in DSH.
  • Reads https://models.dev/api.json; falls back to the models.dev GitHub dev tarball on failure.
  • Caches the directory to the system temporary directory with a validity period of 24 hours.

How it works

After the plugin is enabled, DSH processes the following steps upon startup:

  1. Wait for settings to be available.
  2. Read https://models.dev/api.json.
  3. If the read fails, fall back to the models.dev GitHub dev tarball.
  4. Match models under llm-pi-ai and fill in missing reasoning tiers.
  5. Cache the directory to the system temporary directory with a validity period of 24 hours.

This plugin has no plugin configuration items and no prepare installation script.

Installation and Enabling

You need dsh, pnpm, and git available before installing.

This package requires a Node.js version satisfying:

node ^22.19.0 || >=24.0.0

It is zero-build JavaScript, so you don’t need to add pnpm allowBuilds for it.

The following uses the web profile to install:

dsh plugin --profile web add github:aerince/dsh-models-dev-reasoning

If you need to install a fixed version:

dsh plugin --profile web add github:aerince/dsh-models-dev-reasoning#v0.1.0

If you need to install a fixed commit:

dsh plugin --profile web add github:aerince/dsh-models-dev-reasoning#0893e91cbe7bdc684eaf1ed52767b8ca3289c751

Restart DSH after installation.

DSH will write the plugin to $DSH_HOME/profiles/<profile>; please ensure that the CLI and Desktop use the same DSH home.

Verification and Uninstallation

After installation, check the configuration first:

dsh --profile web --dump-config

The models-dev-reasoning should appear in the configuration tree.

Then open a third-party or custom model that previously did not have reasoningEfforts and check if the reasoning tiers appear in the model selector; models that already have reasoningEfforts should remain unchanged.

If you do not need this plugin, you can remove it:

dsh plugin --profile web remove dsh-models-dev-reasoning

Use Cases and Notes

Suitable for people who use DSH to manage third-party or custom models and want to uniformly display reasoning tiers.

Notes when using:

  • The plugin will only write missing reasoningEfforts under llm-pi-ai and will not overwrite existing configurations.
  • The plugin will access models.dev and the GitHub fallback address.
  • The plugin will write optional cache to the system temporary directory; failure to write the cache will not prevent the plugin from running.
  • The plugin runs with the permissions of the current dsh process; you should check the source code and license before installing.

Getting

  • GitHub: dsh-models-dev-reasoning
  • Plugin directory: You can look for the plugin by name dsh-models-dev-reasoning in the DSH plugin directory