Foreword

DeepSeek Harness (DSH) adopts the “everything is a plugin” philosophy, aiming to extend model inference capabilities through plugins. When building agents or handling localized tasks, developers often need to avoid network dependencies. The dsh-variance plugin provides a solution for such needs.

Plugin Introduction

dsh-variance is a pure Node.js implementation of a model inference plugin maintained by uckkk. It does not depend on any network requests and focuses on local computation.

Core Features

  1. Pure Node.js implementation: Runs on Node.js with no external network calls.
  2. Tool registration: After installation, the plugin registers corresponding tools in the DSH session, allowing developers to invoke them directly.

Installation and Enablement

Install the plugin using the official command:

dsh plugin add github:uckkk/dsh-variance

After installation, you can call the tools registered by the plugin in the session.

Use Cases and Precautions

  • Suitable for model inference tasks that need to run locally and do not involve external API calls.
  • The plugin runs with the permissions of the current DSH process.
  • It is recommended to check the source code and license (MIT) before installation.
  • Dependencies include @deepseek-ai/cordis (^4.0.1) and @deepseek-ai/dsh-tools (>=0.1.0-rc.6).

References