Preface

WorkBuddy desktop clients are logged into models (DeepSeek / GLM / Kimi / MiniMax, etc.), but do not expose a public API. Existing local solutions either rely on undocumented reverse-engineering protocols or require running additional processes. This plugin listens on 127.0.0.1 and converts the models already logged into the local WorkBuddy desktop client into locally accessible OpenAI- and Anthropic-compatible interfaces, for use by DeepSeek Harness (DSH) or other clients that support custom Base URLs.

Core Features

  1. Interface Conversion: Converts locally logged-in WorkBuddy models into OpenAI / Anthropic-compatible interfaces.
  2. Zero Dependencies: Listens only on 127.0.0.1 with no runtime dependencies.
  3. Console: Includes a web console (Chinese/English bilingual) for status monitoring, model registration, and diagnostics.
  4. Native Integration: Provided as a native plugin for DeepSeek Harness (DSH).
  5. Multi-Client Support: Supports integration with Claude Code / opencode / Cursor / Trae / Cherry Studio / LobeChat, and others.

Installation and Enablement

First install the plugin using the official command, then complete model registration in DSH settings.

dsh plugin --profile desktop add github:Ianzhyh/workbuddy-to-dsh

After installation, restart DeepSeek Harness, complete model registration in the WorkBuddy section of the settings page, and save.

Notes

  • Unofficial Approach: Depends on WorkBuddy’s undocumented login credential storage format; use at your own risk.
  • Single-Account Driven: Only reads credentials for the locally logged-in account; it does not provide multi-account pooling or proxy calling services.
  • Protocol Risk: The upstream protocol is not documented; availability is not guaranteed, so it is recommended to use it as-is.
  • Security Isolation: Listens only on 127.0.0.1, is not exposed externally, and does not embed any keys.
  • License: Open-sourced under the MIT license.

Postscript

This plugin is suitable for developers who need to leverage WorkBuddy account resources locally. By integrating as a DSH plugin, it avoids the need to maintain an additional bridge process and can be used directly in the DSH model selector.