Introduction

DeepSeek Harness (DSH) adopts a plugin-based architecture. When managing providers and models, the built-in UI often cannot meet fine-grained configuration requirements. The dsh-model-cap plugin adds a dedicated settings navigation item, “Provider Configuration”, for handling CRUD operations on configuration files and advanced declarations.

Core Features

Provider Management

Supports adding, editing, and deleting providers. You can configure the provider address, protocol (openai-completions or openai-responses), and API Key (blank is supported, which is useful for local models or no-authentication gateways).

Model Management

Supports adding, editing, deleting, and bulk deleting models. Provides drag-to-reorder (with a blue drop indicator line), search, and the ability to pull models from a gateway.

Reasoning and Image Understanding Declarations

Supports declaring reasoning levels reasoningEfforts (supports minimal to max; leaving it blank means off). Supports declaring image understanding capability as input: [text, image].

Connectivity and Maintenance

Provides interfaces for testing a single provider and testing all providers. Supports temporarily disabling specific providers, archiving everything, and re-enabling them at any time.

Runtime Behavior

The plugin writes directly to the llm-pi-ai namespace, and changes take effect immediately after saving. If a session is interrupted, the system automatically recovers based on the revision barrier to prevent configuration loss.

Installation and Configuration

Install Command

dsh plugin --profile web add dsh-model-cap

Local Development

If you need to develop from source locally:

dsh plugin --profile web add file:/absolute/path/to/dsh-model-cap

Environment Requirements

  • DeepSeek Harness: >= 0.2.0-rc.1
  • Node.js: >= 22

Self-Contained Notes

This plugin does not depend on dsh-react-kit. Template components have been inlined into lib/client.js via the build script, and the source snapshot is located in vendor/dpk/. This means it can run as a single package installation without relying on an external component library.

Summary

dsh-model-cap focuses on solving the cumbersome provider and model configuration tasks in DSH, ensuring configuration stability through immediate hot effects and a revision barrier mechanism. It is suitable for developers who need fine-grained management of multiple OpenAI-compatible endpoints or local models.

Related Links:
- GitHub Repository
- Plugin Directory