Introduction

DeepSeek Harness (DSH) adopts a plugin-based architecture. When using third-party models, developers often need to manually manage API keys or spend time on model selection and protocol adaptation (such as Anthropic Prompt Cache). The zenmux/dsh-plugins plugin addresses this pain point: it allows browser OAuth login, direct access to ZenMux’s model library, and automatically selects the optimal protocol based on model capabilities.

Plugin Positioning

This is a DSH plugin maintained by ZenMux. Its core function is to integrate ZenMux models into DSH, supporting browser OAuth login without manually copying API keys. The plugin includes 150 searchable text models and automatically routes to the Anthropic Messages protocol (with Prompt Cache support) or the OpenAI Chat Completions protocol based on model capabilities.

Core Features

  • Browser OAuth Login: Supports completing authorization in a browser; tokens can refresh automatically.
  • Model Search: Built-in 150 text models, searchable directly in the Web interface by name or ID.
  • Protocol-aware routing:
    • 126 models are routed through Anthropic Messages, supporting Prompt Cache control.
    • 24 models are routed through OpenAI Chat Completions.
  • Session association: Each ZenMux inference request carries the current DSH session ID (x-zenmux-session-id) in the request headers.
  • Command-line interaction: Provides commands such as /zenmux login, /zenmux status, /zenmux logout, and /zenmux models refresh.

Installation and Enablement

Installing this plugin requires updating it through the DSH command line. Ensure that DeepSeek Harness is installed in the current environment.

dsh plugin --profile web update @zenmux/dsh-plugins --latest
dsh web

After running the commands above, start the DSH Web interface, and the plugin will be available.

Typical Usage

  1. Login: Enter /zenmux login in the DSH chat window and complete authorization in the browser as prompted.
  2. Status check: Use /zenmux status to confirm the connection status.
  3. Logout: If you need to disconnect, enter /zenmux logout.
  4. Model selection: Open the model selector in the Web interface and directly enter a model name or ID to search. Based on category, models are divided into “ZenMux · Anthropic” and “ZenMux · OpenAI”.

Updates and Maintenance

Updating the plugin requires stopping the current DSH process, running the update command, and then restarting:

dsh plugin --profile web add @zenmux/dsh-plugins@latest
dsh web

If you need to refresh the model list (pulling the latest data from ZenMux), you can run /zenmux models refresh in a session. This command updates the configuration file; it is recommended to restart DSH afterwards.

Notes

  • Manual list override: If the plugin shows fewer models than expected, an older local manual list may be overriding the built-in catalog. Edit ~/.dsh/settings.yaml, remove the models: arrays under llm-pi-ai.providers.zenmux and llm-pi-ai.providers.zenmux-models, and restart DSH.
  • Authentication errors: If you encounter API key is invalid or AUTH errors, rerun /zenmux login.
  • Network and TLS: OAuth requests connect directly by default; explicit HTTP(S) or SOCKS proxies can be configured. If using custom TLS interception, set the environment variable NODE_EXTRA_CA_CERTS before starting DSH.
  • License: This plugin follows the MIT License.

Conclusion

This plugin provides DSH users with a model access solution that does not require manually maintaining keys and optimizes the use of Prompt Cache. The DeepSeek Harness ecosystem is community-driven, and the plugin catalog (such as SkillHub) has no official affiliation with DeepSeek or High-Flyer. For more details and source code, please visit its GitHub repository.