AG (Antigravity) Local Bridge

Exposes your running Antigravity instance as a local OpenAI-compatible API on localhost:11435.

Use your Antigravity subscription with any tool that speaks OpenAI — opencode, aider, continue.dev, or plain curl.

How it Works

Your tool → HTTP :11435 → VS Code extension → Antigravity sidecar (ConnectRPC) → Cloud AI

The extension runs inside Antigravity's VS Code process, discovers the sidecar via process inspection, intercepts CSRF tokens from Antigravity's own traffic, and proxies your requests through the same authenticated channel Antigravity uses internally.

Features

• OpenAI-compatible API — drop-in replacement for any tool expecting OpenAI format
• Multimedia input support — images, audio, video, and files from OpenAI/Responses clients are forwarded through Antigravity's media channel when possible
• Workspace-aware — automatically detects and sets the correct project context via x-workspace-dir / x-workspace-uri headers
• Conversation multiplexing — reuses Cascade conversations for efficiency, with automatic retry on capacity errors
• Streaming & non-streaming — both modes supported

Available Models

Installation

From Open VSX (recommended)

1. Open Antigravity
2. Go to Extensions (Ctrl+Shift+X)
3. Search for "AG Local Bridge" by marcodiniz
4. Click Install
5. Reload Antigravity (Ctrl+Shift+P → Developer: Reload Window)

Manual install

1. Clone into your Antigravity extensions directory:

   # Windows
   git clone https://github.com/marcodiniz/ag-local-bridge "%USERPROFILE%\.antigravity\extensions\ag-local-bridge-1.0.0-universal"
   
   # macOS / Linux
   git clone https://github.com/marcodiniz/ag-local-bridge ~/.antigravity/extensions/ag-local-bridge-1.0.0-universal
   

2. Reload Antigravity (Ctrl+Shift+P → Developer: Reload Window)

Verify

Look for "AG Local Bridge" in the Output panel — you should see:

✅ Server running on http://localhost:11435

Usage

With opencode

Add to ~/.config/opencode/opencode.json:

{
  "provider": {
    "ag-local-bridge": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "AG Local Bridge",
      "options": {
        "baseURL": "http://localhost:11435/v1",
        "apiKey": "local"
      },
      "models": {
        "antigravity-claude-sonnet-4-6": {
          "name": "Claude Sonnet 4.6 (Antigravity)",
          "modalities": { "input": ["text", "image", "audio", "video", "file"], "output": ["text"] },
          "limit": { "context": 200000, "output": 64000 }
        },
        "antigravity-claude-opus-4-6-thinking": {
          "name": "Claude Opus 4.6 Thinking (Antigravity)",
          "modalities": { "input": ["text", "image", "audio", "video", "file"], "output": ["text"] },
          "limit": { "context": 200000, "output": 64000 }
        },
        "antigravity-gemini-3.1-pro-high": {
          "name": "Gemini 3.1 Pro High (Antigravity)",
          "modalities": { "input": ["text", "image", "audio", "video", "file"], "output": ["text"] },
          "limit": { "context": 1048576, "output": 65535 }
        },
        "antigravity-gemini-3.1-pro-low": {
          "name": "Gemini 3.1 Pro Low (Antigravity)",
          "modalities": { "input": ["text", "image", "audio", "video", "file"], "output": ["text"] },
          "limit": { "context": 1048576, "output": 65535 }
        },
        "antigravity-gemini-3-flash": {
          "name": "Gemini 3 Flash (Antigravity)",
          "modalities": { "input": ["text", "image", "audio", "video", "file"], "output": ["text"] },
          "limit": { "context": 1048576, "output": 65536 }
        },
        "antigravity-gpt-oss-120b": {
          "name": "GPT-OSS 120B Medium (Antigravity)",
          "modalities": { "input": ["text", "image", "audio", "video", "file"], "output": ["text"] },
          "limit": { "context": 128000, "output": 16384 }
        }
      }
    }
  }
}

Then select ag-local-bridge/antigravity-claude-sonnet-4-6 as your model.

Media support: The modalities field enables image/audio/video/file input in clients that expose those attachments. Text-only requests use the faster raw inference path; requests with media use the Antigravity Cascade media channel.

With curl

# List models
curl http://localhost:11435/v1/models

# Chat completion
curl http://localhost:11435/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "antigravity-claude-sonnet-4-6",
    "messages": [{"role": "user", "content": "Hello!"}],
    "stream": false
  }'

# Streaming
curl http://localhost:11435/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "antigravity-gemini-3.1-pro-high",
    "messages": [{"role": "user", "content": "Hello!"}],
    "stream": true
  }'

# With image (base64 data URL)
curl http://localhost:11435/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "antigravity-claude-sonnet-4-6",
    "messages": [{"role": "user", "content": [
      {"type": "text", "text": "What do you see?"},
      {"type": "image_url", "image_url": {"url": "data:image/png;base64,iVBOR..."}}
    ]}],
    "stream": false
  }'

# Responses API
curl http://localhost:11435/v1/responses \
  -H "Content-Type: application/json" \
  -d '{
    "model": "antigravity-claude-sonnet-4-6",
    "input": "Hello from the Responses API"
  }'

With any OpenAI-compatible client

Base URL: http://localhost:11435/v1
API Key:  anything (not validated)

API Endpoints

Multimedia Support

Media sent via OpenAI chat content parts or Responses input items is handled as follows:

1. Base64 data URLs (data:image/png;base64,..., data:audio/wav;base64,..., etc.) — decoded and forwarded inline
2. Remote URLs (https://...) — downloaded, then forwarded inline
3. File URIs (file:///C:/path/to/media.mp4) — read directly from disk
4. Responses items (input_image, input_audio, input_file) — converted to the same media channel

The sidecar ultimately decides which model/media combinations are accepted. Unsupported media types will surface as upstream provider errors.

Workspace Context

Pass workspace context via HTTP headers:

When set, the bridge switches the active VS Code workspace folder before creating a Cascade, ensuring the agent operates in the correct project context.

Architecture

The extension uses a 2-tier fallback strategy:

1. Tier 1 — Sidecar ConnectRPC (preferred): Discovers the Antigravity sidecar process, connects via HTTP/2 with CSRF authentication, and uses the Cascade API (StartCascade → SendUserCascadeMessage → poll GetCascadeTrajectory). Conversations are multiplexed and the active workspace is auto-detected.

2. Tier 2 — Command Dispatch: Last resort — fires the message through antigravity.executeCascadeAction.

Commands

Configuration

Requirements

• Antigravity installed and running
• Active Antigravity subscription (Free/Pro/Teams/Enterprise)

Contributing

We use a beta → master release flow:

1. Fork & branch off beta (the default branch).
2. Open your PR against beta — GitHub will target it automatically.
3. Once merged, a pre-release is published to Open VSX for beta testers.
4. When a beta build is ready for general availability, the maintainer merges beta → master to cut a stable release.

Note: master is protected — direct pushes are not allowed. All changes flow through beta first.

License

MIT