Skip to content

antigravity.md

Latest commit

 

History

History
382 lines (308 loc) · 9.75 KB

antigravity.md

File metadata and controls

382 lines (308 loc) · 9.75 KB
Raw

Antigravity Provider Setup Guide

Complete guide for configuring Antigravity provider with oh-my-opencode-slim to access Claude and Gemini models through the "google" provider configuration.

Table of Contents

Overview

Antigravity provides access to high-quality Claude and Gemini models through a unified interface. This guide shows you how to configure oh-my-opencode-slim to use Antigravity via the "google" provider configuration.

Benefits:

  • Access to Claude Opus 4.5 Thinking and Sonnet 4.5 Thinking models
  • Gemini 3 Pro High and Flash models
  • Unified configuration through OpenCode
  • Excellent for complex reasoning tasks (Claude) and fast responses (Gemini)

Installation

Prerequisites

  1. Install oh-my-opencode-slim:

    bunx oh-my-opencode-slim@latest install

  2. Set up Antigravity/LLM-Mux: Follow the installation guide at https://nghyane.github.io/llm-mux/#/installation

  3. Start the Antigravity service: Ensure your Antigravity service is running on http://127.0.0.1:8317

Configuration

Step 1: Configure Provider in OpenCode

Edit ~/.config/opencode/opencode.json and add the "google" provider configuration:

{
  "provider": {
    "google": {
      "options": {
        "baseURL": "http://127.0.0.1:8317/v1beta",
        "apiKey": "sk-dummy"
      },
      "models": {
        "gemini-3-pro-high": {
          "name": "Gemini 3 Pro High",
          "attachment": true,
          "limit": {
            "context": 1048576,
            "output": 65535
          },
          "modalities": {
            "input": ["text", "image", "pdf"],
            "output": ["text"]
          }
        },
        "gemini-3-flash": {
          "name": "Gemini 3 Flash",
          "attachment": true,
          "limit": {
            "context": 1048576,
            "output": 65536
          },
          "modalities": {
            "input": ["text", "image", "pdf"],
            "output": ["text"]
          }
        },
        "claude-opus-4-5-thinking": {
          "name": "Claude Opus 4.5 Thinking",
          "attachment": true,
          "limit": {
            "context": 200000,
            "output": 32000
          },
          "modalities": {
            "input": ["text", "image", "pdf"],
            "output": ["text"]
          }
        },
        "claude-sonnet-4-5-thinking": {
          "name": "Claude Sonnet 4.5 Thinking",
          "attachment": true,
          "limit": {
            "context": 200000,
            "output": 32000
          },
          "modalities": {
            "input": ["text", "image", "pdf"],
            "output": ["text"]
          }
        }
      }
    }
  }
}

Step 2: Configure Agent Models

Edit ~/.config/opencode/oh-my-opencode-slim.json and add the Antigravity preset:

{
  "preset": "antigravity",
  "presets": {
    "antigravity": {
      "orchestrator": {
        "model": "google/claude-opus-4-5-thinking",
        "skills": ["*"],
        "mcps": ["websearch"]
      },
      "oracle": {
        "model": "google/gemini-3-pro-high",
        "variant": "high",
        "skills": [],
        "mcps": []
      },
      "librarian": {
        "model": "google/gemini-3-flash",
        "variant": "low",
        "skills": [],
        "mcps": ["websearch", "context7", "grep_app"]
      },
      "explorer": {
        "model": "google/gemini-3-flash",
        "variant": "low",
        "skills": [],
        "mcps": []
      },
      "designer": {
        "model": "google/gemini-3-flash",
        "variant": "medium",
        "skills": ["agent-browser"],
        "mcps": []
      },
      "fixer": {
        "model": "google/gemini-3-flash",
        "variant": "low",
        "skills": [],
        "mcps": []
      }
    }
  }
}

Provider Setup

Available Models

Model Type Context Window Output Limit Best For
claude-opus-4-5-thinking Claude 200K tokens 32K tokens Complex reasoning, orchestrator
claude-sonnet-4-5-thinking Claude 200K tokens 32K tokens Balanced reasoning and speed
gemini-3-pro-high Gemini 1M tokens 65K tokens High-quality responses
gemini-3-flash Gemini 1M tokens 65K tokens Fast responses, cost-effective

Model Recommendations

For Orchestrator: Use claude-opus-4-5-thinking

  • Best reasoning capabilities for multi-agent coordination
  • Excellent at planning complex workflows
  • Can handle large context windows

For Oracle: Use gemini-3-pro-high

  • Strategic thinking and debugging
  • High-quality architectural advice

For Support Agents: Use gemini-3-flash

  • Fast and cost-effective for routine tasks
  • Good for librarian, explorer, designer, and fixer roles

Agent Model Assignment

Basic Configuration

The preset assigns models based on their strengths:

{
  "presets": {
    "antigravity": {
      "orchestrator": { "model": "google/claude-opus-4-5-thinking" },
      "oracle": { "model": "google/gemini-3-pro-high", "variant": "high" },
      "librarian": { "model": "google/gemini-3-flash", "variant": "low" },
      "explorer": { "model": "google/gemini-3-flash", "variant": "low" },
      "designer": { "model": "google/gemini-3-flash", "variant": "medium" },
      "fixer": { "model": "google/gemini-3-flash", "variant": "low" }
    }
  }
}

Custom Assignments

You can customize which model each agent uses:

Example: All Claude for reasoning-heavy work:

{
  "presets": {
    "claude-heavy": {
      "orchestrator": { "model": "google/claude-opus-4-5-thinking" },
      "oracle": { "model": "google/claude-sonnet-4-5-thinking", "variant": "high" },
      "librarian": { "model": "google/claude-sonnet-4-5-thinking", "variant": "low" },
      "explorer": { "model": "google/claude-sonnet-4-5-thinking", "variant": "low" },
      "designer": { "model": "google/claude-sonnet-4-5-thinking", "variant": "medium" },
      "fixer": { "model": "google/claude-sonnet-4-5-thinking", "variant": "low" }
    }
  }
}

Example: Mixed setup with cost optimization:

{
  "presets": {
    "cost-optimized": {
      "orchestrator": { "model": "google/claude-sonnet-4-5-thinking" },
      "oracle": { "model": "google/gemini-3-pro-high", "variant": "high" },
      "librarian": { "model": "google/gemini-3-flash", "variant": "low" },
      "explorer": { "model": "google/gemini-3-flash", "variant": "low" },
      "designer": { "model": "google/gemini-3-flash", "variant": "low" },
      "fixer": { "model": "google/gemini-3-flash", "variant": "low" }
    }
  }
}

Usage Examples

Switching to Antigravity Preset

# Edit the preset in your config
export OH_MY_OPENCODE_SLIM_PRESET=antigravity
opencode

Testing Configuration

  1. Verify provider connection:

    opencode auth status

  2. Test agent responses:

    opencode
# In OpenCode, run: ping all agents

  3. Check logs for any connection issues:

    tail -f ~/.config/opencode/logs/opencode.log

Example Workflow

Once configured, you can ask the orchestrator to handle complex tasks:

Orchestrator, please analyze this codebase and create a comprehensive refactoring plan.
Use your team to:
1. Map the current architecture
2. Identify bottlenecks and issues  
3. Create a detailed implementation plan
4. Estimate effort and risks

The orchestrator will delegate to specialists using the configured Antigravity models.

Troubleshooting

Provider Connection Issues

Problem: Agents not responding or showing connection errors

Solutions:

  1. Verify Antigravity service is running:

    curl http://127.0.0.1:8317/v1beta/models

  2. Check the baseURL in your config matches your service:

    {
  "options": {
    "baseURL": "http://127.0.0.1:8317/v1beta"
  }
}

  3. Verify your API key (can be dummy value like "sk-dummy")

Model Not Found Errors

Problem: "Model not found" or similar errors

Solutions:

  1. Ensure the model names match exactly:

    • claude-opus-4-5-thinking (not claude-opus-4.5-thinking)
    • gemini-3-flash (not gemini-3-flash-preview)

  2. Check your provider configuration includes all required fields:

    {
  "models": {
    "model-name": {
      "name": "Display Name",
      "attachment": true,
      "limit": { "context": 1048576, "output": 65535 },
      "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
    }
  }
}

Authentication Issues

Problem: 401 Unauthorized or similar auth errors

Solutions:

  1. The apiKey can be any dummy value for Antigravity
  2. Ensure the baseURL is correct and accessible
  3. Restart OpenCode after changing configuration

Performance Issues

Problem: Slow responses or timeouts

Solutions:

  1. Use gemini-3-flash for faster responses on support agents
  2. Check your Antigravity service performance
  3. Consider reducing context sizes for large files

Switching Between Presets

To test different configurations:

# Use environment variable for quick testing
export OH_MY_OPENCODE_SLIM_PRESET=openai
opencode

# Or edit ~/.config/opencode/oh-my-opencode-slim.json
# Change the "preset" field and restart OpenCode

Additional Resources