The core design philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” During use, obtaining and managing API keys for various models is a common operational burden. dsh-our-free-model is a DeepSeek Harness plugin that aims to provide a solution for using frontier models without account registration or API key entry.

The plugin is maintained by developer Ebony-Vinyl and belongs to the “Model Inference” category. It can access frontier models, including DeepSeek V4.1 Flash and Kimi K3, without login, registration, or any API key. The project is open source under the MIT license and has accumulated 14,626 Stars.

Core Features

  • Zero dependencies, no build steps: Pure plugin mounting, no kernel modification, ready to use after installation.
  • Out of the box: Built-in access to thirteen free channels (including CodeArts, Tencent CodeBuddy, Alibaba Qoder, ByteDance TRAE, iFlytek Loomy, etc.), plus the Kilo free lane and the original anonymous free channel.
  • Intelligent model selection: The selector only broadcasts models available at the current network egress; unavailable models are removed from the selector and only retained as records on the settings page.
  • Thinking intensity control: Supports Light / Balanced / Deep levels, corresponding to different output token budgets.
  • Nine-page frosted glass interface: Provides nine settings pages, including model list, data dashboard, runtime logs, gateway settings, aggregated channels, token savings, etc.
  • In-app upgrade and hot reload: Supports one-click upgrade (with verification and rollback) and immediate activation of code changes without restarting the application.
  • Local gateway and Anthropic interface: Provides an OpenAI-compatible local forwarding port and supports the Anthropic interface.
  • LAN forwarding relay: Provides an independent forwarding gate and supports LAN access.

Installation and Enabling

  1. Download the plugin files to a local directory.
  2. Execute the installation command in the terminal (replace with the actual path):
    dsh plugin --profile web add /绝对路径/dsh-our-free-model
  1. Restart the application after installation. --profile should be set to the actual profile name in use.

Typical Usage

  • Configure channels: On the “Free model access” or “Aggregated channels” settings page, the system automatically loads available free models. For aggregated channels, you can add third-party endpoints (such as OpenRouter, Groq, etc.) and enter a local key.
  • Select models: After the plugin starts, the model selector automatically updates based on network egress availability. Models whose routing is rejected upstream are removed from the selector but retained in the settings page records.
  • Local invocation: Enable the local gateway via the settings page; other tools can call these models through the specified Base URL and key.

Notes

  • Thinking cost: For models where thinking cannot be disabled (e.g., MiMo V2.6), the total output budget for the three thinking intensity levels is doubled (4,096 / 16,384 / model limit), because thinking and main content share the output quota.
  • Privacy and security: Usage data remains entirely on the local machine and is not uploaded. The Kilo pool upstream records prompts; do not send sensitive content. Keys are only written to a local settings file with 0600 permissions.
  • Desktop client restrictions: The desktop profile gate restricts how the plugin can be installed; refer to the plugin instructions for specific requirements.

Summary

dsh-our-free-model aggregates multiple free channels to alleviate the pain point for DeepSeek Harness users in obtaining API keys, while providing robust local gateway and interface management features.

If you need to view the complete documentation, source code, or submit feedback, please visit the GitHub repository.