Introduction

The plugin-based architecture of DeepSeek Harness (DSH) allows developers to extend model routing capabilities through configuration. When integrating Ant Digital MaaS or NVIDIA NIM, manually configuring adapters or adding extra HTTP proxies is often required. The dsh-cloud-model-providers plugin reuses DSH’s built-in llm-pi-ai adapter to provide streaming model routing access to both types of services while maintaining unified credential management and streaming logic.

Plugin Purpose

Plugin Name: bitdg/dsh-cloud-model-providers
Maintainer: BitDG
License: MIT
Core Functionality: Adds model routing for Ant Digital MaaS and NVIDIA NIM to DSH web or headless profiles.

Core Capabilities

  1. Model Routing Support: Provides ant-maas-responses, ant-maas-chat, and nvidia routing.
  2. Protocol Compatibility: Supports OpenAI Responses over SSE and Chat Completions over SSE protocols.
  3. Feature Completeness: Compatible with DSH’s llm-pi-ai adapter, with native support for streaming, tool calling, usage statistics, and request cancellation.
  4. Credential Management: Does not store API keys; they must be provided via environment variables or DSH settings.

Installation and Enablement

The plugin works as a declarative configuration layer (cordis.patch.yml) and becomes active after installation.

  1. Install the plugin:
    dsh plugin --profile web add github:BitDG/dsh-cloud-model-providers

This command installs the plugin into the `web` profile and activates its declared configuration layer.
  1. Verify installation:
    Before starting the application, check whether the configuration has taken effect:
    dsh --profile web --dump-config

The output should include the `# == @deepseek-ai/dsh-base, patched by dsh-cloud-model-providers` marker and the modified `llm-pi-ai` configuration lines.
  1. Uninstall the plugin:
    dsh plugin --profile web remove dsh-cloud-model-providers

Configuration and Usage

The plugin requires credentials through environment variables and appropriate routing and model selection in configuration.

  1. Set environment variables:
    Set the API keys before starting DSH:
    export MAAS_API_KEY='your-maas-key'
    export NVIDIA_API_KEY='your-nvidia-key'

Or enter them directly on the "Model Settings" page in the Web UI.

  1. Select models:
    In DSH’s model list, select the corresponding route based on the protocol:

    • ant-maas-responses: For OpenAI Responses over SSE protocol, with default model lingdt-3.0-flash.
    • ant-maas-chat: For Chat Completions over SSE protocol, with default model qwen3.6-plus.
    • nvidia: For Chat Completions over SSE, where the model source depends on the installed pi-ai directory in DSH.

    If using Headless mode, specify the default model in cordis.patch.yml:

    - id: agent-default-model
      config:
        provider: ant-maas-responses
        model: lingdt-3.0-flash

Notes

  • No container deployment: The plugin acts only as a network client calling official APIs and does not deploy self-hosted NVIDIA NIM containers.
  • Model list: The preconfigured MaaS model list is small. Add other models in DSH settings and confirm protocol support yourself.
  • Protocol changes: MaaS model availability and protocol support may change according to official documentation. Consult the official catalog before use.

Summary

This plugin extends DSH’s model integration capabilities through declarative configuration, enabling developers to seamlessly use the streaming model interfaces of Ant Digital MaaS and NVIDIA NIM. For more technical details and source code, refer to the official repository.