dsh-antigravity

dsh-antigravity

English | 简体中文

Google Antigravity / Cloud Code Assist model provider for DeepSeek Harness.

This is a DSH Web plugin. It registers a DSH LlmAdapter under provider route antigravity, stores OAuth credentials under DSH home, talks to the Cloud Code Assist streaming API directly, and provides full bilingual (English & Simplified Chinese) i18n support in the Web settings page.

Unofficial integration. This project is not affiliated with or endorsed by Google. Use it only with accounts and services you are authorized to access.

Install into DSH Web

Option 1: From npm (Recommended)

dsh plugin --profile web add dsh-antigravity

Option 2: Direct from GitHub

dsh plugin --profile web add github:Miyamiz39/dsh-antigravity

Option 3: From Local Release Tarball

npm run pack:dist
dsh plugin --profile web add ./dist/dsh-antigravity-0.0.9.tgz

The package declares a DSH bundle patch, so installation automatically mounts the host plugin and browser settings page.

If your DSH version does not support dsh plugin add, copy the package into the Web profile manually:

cp -R dsh-antigravity "$DSH_HOME/profiles/web/node_modules/"

Then add the plugin to the profile cordis.patch.yml:

- insert:
    - id: llm-antigravity
      name: dsh-antigravity

Restart DSH:

dsh web

Login

Open Settings > Antigravity and click Login. The settings page starts Google OAuth and refreshes quota after login completes.

After login, the settings page displays account information, quota grouped by model family, reset times, and model selector options:

• Gemini Models — Gemini Flash / Pro variants share one quota pool (green bars).
• Claude and GPT models — Claude Opus, Claude Sonnet, and GPT-OSS share a separate 3P quota pool (cyan bars).

Each group shows a 5-hour limit (smooths short-term demand) and a weekly limit (tied to your subscription tier) with a live countdown to reset.

The plugin starts a loopback OAuth callback server at http://localhost:51121/oauth-callback. If the Web service cannot open a browser, run the terminal helper on the same machine:

node "$DSH_HOME/profiles/web/node_modules/dsh-antigravity/bin/antigravity-login.mjs"

Credentials are stored at:

$DSH_HOME/storages/antigravity-oauth.json

Keep that file private. It contains access and refresh tokens.

Proxy Settings

Open Settings > Antigravity > Proxy Settings to enable a dedicated proxy and enter the local HTTP proxy port (default 7890). When enabled, Google OAuth token, account, quota, model-discovery, and chat-stream requests are routed through http://127.0.0.1:<port>. Changes apply to new requests immediately without restarting DSH.

The Test connection button probes Google OAuth through the port currently entered and reports the HTTP status and latency. This supports local HTTP/HTTPS CONNECT proxies only, not SOCKS ports. The Google authorization page opened in the browser continues to use the browser/system proxy.

The proxy configuration is stored at:

$DSH_HOME/storages/antigravity-proxy.json

Models

After login, select the Antigravity provider in DSH's model picker. Use the model selector in Settings > Antigravity to enable or disable individual models (enabled models are prioritized at the top of the list) — the live remaining quota percentage is shown next to each one.

Registered model IDs:

Model ID	Name	Quota pool
gemini-3.7-flash	Gemini 3.7 Flash	Gemini
gemini-3.6-flash	Gemini 3.6 Flash	Gemini
gemini-3.5-flash	Gemini 3.5 Flash	Gemini
gemini-3.1-pro	Gemini 3.1 Pro	Gemini
gemini-3.1-flash-image	Gemini 3.1 Flash Image	Gemini
gemini-3-flash	Gemini 3 Flash	Gemini
gemini-2.5-pro	Gemini 2.5 Pro	Gemini
gemini-2.5-flash	Gemini 2.5 Flash	Gemini
claude-opus-4-6	Claude Opus 4.6	Claude & GPT (3P)
claude-sonnet-4-6	Claude Sonnet 4.6	Claude & GPT (3P)
gpt-oss-120b	GPT-OSS 120B	Claude & GPT (3P)

Models in the same quota pool share a weekly limit and a 5-hour limit. Quota is consumed proportionally to token cost, so heavier models (e.g. Claude Opus) drain the pool faster than lighter ones.

The plugin resolves these public IDs to runtime model IDs using the live fetchAvailableModels catalog when available, with static routing fallbacks.

For reasoning-capable models, the plugin defaults to medium effort (or high when that is the only advertised level), explicitly requests upstream thoughts, and maps streamed thoughts to DSH reasoning content blocks. The effort remains selectable from the model picker.

License

MIT