Introduction

DeepSeek Harness (DSH) adopts a pluggable architecture, allowing users to integrate different model sources in a single interface. For users with existing paid subscriptions, logging in directly with their existing account is usually more convenient than configuring API keys. The dsh-plugin-oauth-subs plugin achieves this by converting subscriptions to services such as ChatGPT Codex, xAI Grok, Zhipu GLM, AWS Kiro, Google Antigravity, and Cursor into models that can be invoked in DSH.

Plugin Overview

  • Name: dsh-plugin-oauth-subs
  • Core Value: Calling subscriptions to multiple AI services inside DeepSeek Harness
  • Maintainer: xxww0098
  • Category: Model inference
  • License: MIT
  • Stars: 4

Core Features

  • Multi-service Login: Supports OAuth login for multiple services, including ChatGPT/Codex, xAI Grok, Zhipu GLM, AWS Kiro, Google Antigravity, and Cursor.
  • Authentication Methods: Supports PKCE and Device Code authentication.
  • Local Proxy: Provides a local Responses proxy (Loopback Proxy).
  • Data Sync: Supports llm-pi-ai synchronization.
  • Management and Tracking: Provides quota management and usage tracking (tokens, cache, call counts, latency).
  • Model Configuration: Supports switching by model family and adjusting the input window.

Installation and Activation

In DeepSeek Harness’s Web environment (--profile web), install the plugin using the following command:

dsh plugin --profile web add https://github.com/xxww0098/dsh-plugin-oauth-subs

After installation, restart the app to make it take effect.

Typical Usage

  1. Open the Workbench: In the DSH sidebar, click Plugins → Subscriptions.
  2. Add an Account: On the Subscriptions page, select a service to sign in, or directly import an already signed-in local state.
  3. Use Models: In the Models tab, enable the desired model families, then select the corresponding model in a Harness conversation.

Notes

  • Data Privacy: Accounts and data are stored only in local DSH configuration files, and traffic is directed only to signed-in services.
  • Cost Estimation: Costs displayed by the plugin are estimates based on public prices, not actual bills.
  • Fast Models: Models with a -fast suffix consume more quota.
  • Reasoning Intensity: The reasoning intensity of a model should be configured in the Harness conversation menu, not in the Models tab.
  • Automatic Updates: Plugin updates are installed automatically, but can also be checked manually in the Settings tab.

Conclusion

This plugin unifies multiple external subscriptions into the DeepSeek Harness workbench, making it suitable for developers who need to use multiple paid model services simultaneously. More details can be found on the catalog page or the GitHub repository.