Preface

When developing agents using DeepSeek Harness (DSH), it is sometimes necessary to connect to local LLM inference services. Manual connection configuration is cumbersome and error-prone. DSH’s “everything is a plugin” philosophy provides a standardized way to manage these connections.

What It Is

dsh-wsl-llamacpp is a plugin for connecting to the OpenAI-compatible interface of local llama.cpp or Unsloth Desktop (default port 8080). The plugin is maintained by user 173787247 and is designed to address the need to manage local inference flows through DSH.

Core Features

  • Connects to OpenAI-compatible interfaces.
  • Supports llama.cpp / Unsloth Desktop.
  • Provides the llama_status tool for checking model reachability and ID.
  • Provides the llama_chat tool for sending /v1/chat/completions requests.
  • Supports reusing existing GGUF files without reinstalling the toolchain.

Installation and Enablement

The plugin installation command is as follows:

dsh plugin --profile web add github:173787247/dsh-wsl-llamacpp

Note: Before installing, ensure that the DSH version is >= 0.1.2. dsh-wsl-kit is an optional suite and is not included in install.sh.

Configuration Points

The plugin supports controlling behavior through configuration items, including baseUrl, defaultModel, and timeoutMs. It also supports reusing existing model configurations through environment variables:
- DSH_LLAMA_BASE
- DSH_LLAMA_MODEL

Typical Usage

  1. Use llama_status to check whether the model is reachable and to obtain its model ID.
  2. Use llama_chat to send a /v1/chat/completions request.
  3. Set baseUrl, defaultModel, and timeoutMs in the configuration file.

Applicable Scenarios and Notes

This plugin is applicable to scenarios where local llama.cpp or Unsloth Desktop services in a WSL environment need to be integrated within DSH.

Notes:
- The plugin runs with the permissions of the current DSH process.
- Before installation, check the source code and license (MIT).
- The plugin directory is a community site with no official affiliation to DeepSeek or High-Flyer.

Conclusion

This plugin simplifies the process of integrating local LLM services into DSH through a standardized interface. For more details, refer to the Plugin Directory or the GitHub repository.