Preface

In the DeepSeek Harness (DSH) ecosystem, developers often need to configure unofficial or custom model providers. These providers frequently lack key parameters such as reasoning level, context window, output limit, and image modality. Manual configuration is not only cumbersome but also error-prone.

Plugin Introduction

dsh-model-reasoning is a DSH plugin maintained by TikaFlow. It reads data from models.dev and automatically fills in reasoning level, maximum context, output limit, and image modality for models from unofficial providers, resolving configuration gaps.

Core Features

  1. Automatic Parameter Filling
    Automatically fills in missing reasoning level (reasoningEfforts), maximum context (contextWindow), output limit (maxTokens), and image modality (input).
  2. Route-Level Compatibility Configuration
    By default, writes route-level compatibility configuration for all providers with api: openai-completions, disabling the developer role (compat.supportsDeveloperRole: false).
  3. Provider Exclusion
    Supports excluding specific providers through a list, completely skipping fill and update operations for them.
  4. Offline Support
    Built-in cache support allows the plugin to be used offline upon startup.

Installation and Activation

Install via command line:

dsh plugin --profile web add github:TikaFlow/dsh-model-fix

After installation, the plugin takes effect automatically in DSH without additional operations.

Typical Usage

Automatic Activation

The plugin uses built-in cache to automatically fill model parameters at startup. It will then automatically keep the data up to date.

Manual Adjustment

Go to Web Settings → “Models” → adjust parameters at the bottom of the page, or manually edit the tikaflow-model-fix configuration section in settings.yaml.

Excluding Specific Providers

List the provider IDs that you do not want the plugin to handle in the configuration.

Use Cases and Notes

This plugin is suitable for scenarios requiring quick configuration of unofficial model providers. When using it, be aware of the host version dependency (DSH 0.1.2-rc.1).

Notes

  • The data source is models.dev.
  • It depends on DSH host version 0.1.2-rc.1.
  • Model parameters filled at startup are automatically restored from backup (recreated on restart).
  • This plugin has been included in awesome-dsh-plugin.

Summary

This plugin lowers the barrier for integrating unofficial models through automatic parameter filling and compatibility configuration.

Plugin Directory
GitHub Repository