DeepSeek Harness (DSH) Plugin: dsh-vertex-argon

Introduction¶

The plugin ecosystem of DeepSeek Harness (DSH) allows users to extend model integration capabilities. dsh-vertex-argon is a plugin adapter for the Google Gemini 4 Argon model, providing dual-channel access support through Google Vertex AI and the Google Gemini Developer API. This plugin fixes HTTP 400 errors caused by the built-in pi-ai adapter when it incorrectly passes the thinkingBudget parameter, and natively supports three reasoning levels, thought summaries, v1beta1 long-text continuation, and signed multi-turn tool call replay.

Plugin Overview¶

• Name: dsh-vertex-argon
• Owner: higekibaka
• Category: Model inference
• License: MIT
• Summary: Google Vertex AI & Gemini API Dual-Route Gemini 4 Argon Plugin for DeepSeek Harness (DSH)

Installation¶

To install this plugin in a local environment, ensure the DeepSeek Harness version meets the requirements (0.1.7-rc.2, 0.2.0-rc.1, or 0.2.0-rc.2) and the Node.js version is ^22.19.0 or >=24.0.0.

Method 1: Install using the CLI

dsh plugin --profile web add dsh-vertex-argon

Method 2: Install in the DSH chat box

Directly enter the following instruction in a DSH session:

请用 plugin_manager 的 install_bundle 安装 dsh-vertex-argon

Configuration and Usage¶

The plugin supports dual-channel configuration. Users can choose one or configure both according to their needs.

Channel A: Google Vertex AI¶

1. Obtain the service account JSON key file in the Google Cloud console.
2. Set the GOOGLE_APPLICATION_CREDENTIALS system environment variable to point to the JSON file path.
3. Select Vertex Argon → Gemini 4 Argon (Vertex AI) in the DSH model selector.

Channel B: Google Gemini API¶

1. Configure GOOGLE_API_KEY in DSH settings.
2. Select Vertex Argon → Gemini 4 Argon (Gemini API) in the model selector.

Core Features¶

This plugin optimizes parameter handling and long-text generation:

• Native reasoning levels: Supports LOW, MEDIUM, and HIGH reasoning levels instead of using the thinkingBudget parameter, which can easily cause errors.
• Thought summaries: Enables includeThoughts: true, with streaming parsing and display of the chain of thought.
• Long-text continuation: Enables the v1beta1 long decoding continuation mechanism, supporting zero-latency continuation across 64K segments.
• Tool call signing: Keeps provider / model and the thoughtSignature signature consistent across the full chain, ensuring normal multi-turn tool calls.
• Parameter sanitization: Automatically removes invalid sampling parameters such as temperature, topP, and topK.
• Automatic detection: Automatically detects dual-channel configuration without manually modifying YAML configuration files.

Settings Page Configuration¶

Go to DSH Settings -> Vertex Argon to adjust the following configuration items in real time:

Known Compatibility Issues¶

When using a Google channel proxy, this plugin has known compatibility issues with billion-context; wait for an upstream fix or avoid them through configuration.

1. HTTP 400 error: When billion-context appends a compression reminder to a user message, it may cause an abnormal order in parts, triggering a Vertex AI error.
2. Signature fragment filtering: Independent { thoughtSignature } fragments in streaming responses may be filtered by the Google loop adapter, affecting the continuity of long-chain reasoning.

Ecosystem References¶

• GitHub repository: https://github.com/higekibaka/dsh-vertex-argon
• Plugin directory: https://www.skillhub.cn/plugins/higekibaka/dsh-vertex-argon