Introduction

DSH’s plugin architecture follows the philosophy of “everything is a plugin,” and LLM Providers can also be extended as plugins. For developers who already have Google Anti Gravity Coding Plan or Cloud Code Assist benefits, a specific need is to bring models like Gemini, Claude, and GPT-OSS under these accounts into the DeepSeek Harness workspace, without introducing additional IDE instances or background subprocesses.

OpenSaozi/dsh-antigravity is a DSH LLM Provider plugin designed for this scenario.

What is this

OpenSaozi/dsh-antigravity is maintained by OpenSaozi and is licensed under MIT.

Its positioning is to act as the LLM Provider for DeepSeek Harness using the Google Anti Gravity Coding Plan, allowing Harness to directly use models such as Gemini, Claude, and GPT-OSS under the account.

This is an independent integration project and is not affiliated with or endorsed by Google, DeepMind, DeepSeek, or the maintainers of pi-ai.

Core Features

Direct Connection to Cloud Code Assist

The plugin connects to Cloud Code Assist directly via HTTPS, without starting a background agy subprocess, nor does it start an IDE instance.

Independent Adapter and Model Directory

The plugin registers an independent google-antigravity adapter, maintains its own model directory, and is responsible for parsing OAuth.

It only reuses PiAiAdapter to complete Harness message and stream conversion.

Dynamic Model Discovery

The plugin queries the fetchAvailableModels endpoint to display Gemini, Claude, and GPT-OSS model variants.

macOS One-Click Authorization

On macOS, the plugin can automatically verify and import the signed macOS Anti Gravity client configuration, and then open the standard browser OAuth flow.

Non-standard paths can be provided via macosApplicationPath.

Built-in Login Page

The plugin provides the ability to start login, poll status, and log out, and registers a dedicated Anti Gravity settings page on the browser side.

Preserve Native Inference Contracts

The plugin preserves thinking signatures, streaming output, and tool calling contracts.

It does not inject provider system prompts, synthetic messages, or custom headers.

Installation and Enablement

First, select the target DSH profile, then install the specified revision version:

dsh plugin --profile <profile> add github:OpenSaozi/dsh-antigravity#<commit-sha>

Where:

  • <profile> is the target DSH profile name
  • <commit-sha> is the repository revision version

After installation, the plugin will contribute Cordis configuration:

  • id: llm-pi-ai-antigravity
  • displayName: Google Anti Gravity

The configuration will reference two credential items:

GOOGLE_ANTIGRAVITY_OAUTH_CREDENTIAL
GOOGLE_ANTIGRAVITY_OAUTH_CLIENT_CONFIG

The bundled local provider stores documentation in:

$DSH_HOME/.credentials.yaml

Deployment environments can also use other credential backends.

Typical Usage

  1. Install the plugin to the target DSH profile:
dsh plugin --profile <profile> add github:OpenSaozi/dsh-antigravity#<commit-sha>
  1. Start OAuth from the Anti Gravity settings page owned by the plugin.

The account needs to hold active service benefits.

  1. The plugin queries the fetchAvailableModels endpoint to display Gemini, Claude, and GPT-OSS model variants.

  2. Initiate model calls or tool calls in Harness.

If quotas are exhausted, a provider error will be returned during inference.

Applicable Scenarios and Notes

Suitable for the following situations:

  • Google-signed Antigravity IDE.app is already installed on macOS, and authorization is completed via browser OAuth.
  • Google Anti Gravity Coding Plan or Cloud Code Assist related service benefits are already held.
  • Want to connect models under the account to DeepSeek Harness while avoiding starting extra IDE instances or background agy subprocesses via the plugin.

Please note:

  • Google subscription paths go through the internal v1internal Cloud Code Assist endpoint, not the public Gemini API. Changes to the endpoint may affect this plugin.
  • The model directory describes provider capabilities and does not guarantee remaining quotas.
  • Other operating systems need to pre-set credentials via the credential service.
  • Credentials are retained in the Harness host credential service and will not leak to settings, model directories, logs, or frontend responses.
  • The plugin runs with the permissions of the current dsh process. Before installing, it is recommended to check the source code, dependencies, and MIT license to decide whether to install it into your own profile.

Ending

OpenSaozi/dsh-antigravity provides a relatively lightweight DSH LLM Provider integration path: through direct connection via Cloud Code Assist, dynamic model discovery, independent adapter, and built-in settings page, it brings models from the Google Anti Gravity Coding Plan into the Harness workspace.

Plugin Directory Page: https://www.skillhub.cn/plugins/OpenSaozi/dsh-antigravity

GitHub: https://github.com/OpenSaozi/dsh-antigravity