Introduction

When managing custom providers in DeepSeek Harness (DSH), you usually need to manually fill in fields such as context window, output limit, reasoning effort levels, and image capabilities. These settings are easy to omit or enter incorrectly when adding new models, which can prevent models from working properly or limit inference performance. This plugin connects to models.dev and automatically fills in these fields by model name, solving the problem of cumbersome and error-prone configuration.

What It Is

dsh-model-info-fill is a DeepSeek Harness plugin maintained by 11zld22. It uses the knowledge base data from models.dev to automatically populate context, output limit, reasoning effort levels, and image capabilities for models in custom providers. The plugin does not overwrite fields that users have already filled in manually, nor does it create providers based on models.dev because there is no Base URL / API Key.

Core Features

The plugin primarily provides the following capabilities:

  1. Automatic field completion: Automatically fills context window contextWindow, output limit maxTokens, reasoning efforts reasoningEfforts, and image input declaration input from models.dev.
  2. Preservation of manual entries: Fields that have already been filled in manually are not overwritten, and the default reasoning level of a provider that has already set reasoning is not overwritten.
  3. Default value handling: Uses default values when no knowledge base match is found; fields that are unclear in the knowledge base are left empty rather than forced.
  4. Smart level detection: Handles models with unspecified levels and does not invent levels; models not present in the knowledge base fall back to default values.
  5. Provider default level: Sets a default reasoning level (reasoning) for a provider, used as the default reasoning strength for new conversations.
  6. Strict mode: Supports strict mode to check default-level compatibility; it writes a level only if all models in that provider support that level.
  7. Model matching logic: Supports exact matching and comparison after removing prefixes and variant suffixes (such as :free).

Installation and Enablement

Install using the official command:

dsh plugin --profile web add github:11zld22/dsh-model-info-fill

After installation, restart DSH. For the desktop version, replace --profile web with --profile desktop. Open Settings → Models to use the plugin.

Typical Usage

At the bottom of the model settings page, a Model Info Fill row is displayed, including the knowledge base sync time and the number of models.

  1. Update knowledge base: Click the Update button to re-fetch data from models.dev.
  2. Automatic completion: When saving a custom model, the plugin automatically fills in missing fields.
  3. Bulk completion: Click Fill All Missing to immediately complete all current custom models and set a default reasoning level for the provider.
  4. Unmatched section: Adjust default values (such as the default level) in the Unmatched section; newly added models that do not match the knowledge base will use these settings.
  5. Unspecified level section: In the Unspecified Level section, for models that only indicate reasoning without explicit discrete levels, choose to keep, clear, or specify a level.
  6. Confirm legacy data: Click Confirm Keep or Clear Level to handle data that may have been filled in by an older version (such as minimal / low / medium / high filled in before 0.2.0).
  7. Strict mode: Check or uncheck Strict mode in the panel to decide whether to verify that all models support a level before writing the default level.

Notes

  1. Providers are not created: The plugin is only used to complete model fields and does not create providers based on models.dev (no Base URL / API Key).
  2. Knowledge base limitations: Models not found in the knowledge base use default values, and fields that are unclear in the knowledge base are left empty.
  3. No invented levels: It does not invent levels for models that are only marked as supporting reasoning but do not have explicit discrete levels in the knowledge base.
  4. Data compatibility: Supports the DSH 0.1.6 and 0.1.7 settings interface. The cache is automatically refreshed after startup if it is older than 7 days.
  5. License: The plugin is released under the MIT License.

Conclusion

This plugin standardizes the process of entering model information, reducing manual maintenance costs. It is especially suitable for developers who need to manage a large number of custom models. For more details, visit GitHub.