aurarouter-antigravity

Google Antigravity MCP provider server for AuraRouter.

aurarouter-antigravity wraps the official google-antigravity Python SDK, exposing Google Antigravity agentic capabilities and Gemini models as a standard Model Context Protocol (MCP) server.

Architecture Overview

graph TD
    Client[User / Application / MAS] --> Router[AuraRouter ComputeFabric]
    Router --> Stack{v2.1 Routing Stack}
    Stack -->|Select Antigravity| MCP[AuraRouter McpProvider]
    MCP --> FastMCP[aurarouter-antigravity FastMCP Server]
    FastMCP --> Provider[AntigravityProvider Engine]
    Provider --> SDK[google-antigravity SDK: Agent / LocalAgentConfig]
    SDK --> Gateway[Google Gemini / Vertex AI Endpoint]

Features

• Native google-antigravity SDK Integration: Directly utilizes google.antigravity.Agent and LocalAgentConfig with streaming response resolution.
• Model Catalog:
  • gemini-3.8-flash: Default Antigravity model for high-efficiency reasoning and agent workflows.
  • gemini-3.1-pro: Deep reasoning and long-context (2M tokens) agent execution.
  • gemini-3.5-flash: Balanced latency and multimodal reasoning.
  • gemini-2.5-pro & gemini-2.5-flash: Proven production generation.
• AuraRouter v2.1 Telemetry & Error Isolation:
  • Maps RESOURCE_EXHAUSTED / QuotaFailure to AntigravityRateLimitError (rate_limit_exceeded) with retry_after_s.
  • Maps context overflow to AntigravityContextLengthError (context_length_exceeded).
  • Maps 5xx / connection timeouts to AntigravityTransientError (provider_transient_error).
• Structured Outputs: Native support for JSON schemas and json_mode.
• Accurate Telemetry: Tracks prompt tokens, completion tokens, and Time-to-First-Token (TTFT) via monotonic clocks.

Installation

pip install aurarouter-antigravity

To install from source:

cd C:\projects\auracore\aurarouter-antigravity
pip install -e .

Environment Configuration

AuraRouter Integration

In your AuraRouter configuration file (auraconfig.yaml), add Antigravity models:

models:
  gemini-3.8-flash:
    provider: mcp
    provider_name: antigravity
    hosting_tier: cloud
    cost_per_1m_input: 0.15
    cost_per_1m_output: 0.60
    context_limit: 1048576
    capabilities: [code, reasoning, agent, chat, multimodal]

  gemini-3.1-pro:
    provider: mcp
    provider_name: antigravity
    hosting_tier: cloud
    cost_per_1m_input: 1.25
    cost_per_1m_output: 10.00
    context_limit: 2097152
    capabilities: [code, reasoning, agent, chat, complex]

roles:
  lead_architect:
    - gemini-3.1-pro
    - claude-3-5-sonnet-20241022
  fast_coder:
    - gemini-3.8-flash
    - gpt-4o-mini

When installed in the Python environment, AuraRouter automatically discovers antigravity via the aurarouter.providers entry point.

Direct Python Usage

from aurarouter_antigravity.provider import AntigravityProvider

provider = AntigravityProvider()

# Single-shot text generation
result = provider.generate(
    prompt="Explain quantum annealing in simple terms.",
    model="gemini-3.8-flash"
)
print(f"Response: {result['text']}")
print(f"Tokens: {result['input_tokens']} in / {result['output_tokens']} out (TTFT: {result['ttft_s']:.3f}s)")

# Multi-turn conversation
history_result = provider.generate_with_history(
    messages=[
        {"role": "user", "content": "Hello!"},
        {"role": "assistant", "content": "Greetings! How may I assist you?"},
        {"role": "user", "content": "What is the capital of France?"}
    ],
    system_prompt="Be concise and factual.",
    model="gemini-3.8-flash"
)
print(history_result["text"])

MCP Tools Exposed

Running Standalone

Start the FastMCP server over standard I/O:

python -m aurarouter_antigravity

Display CLI options and help:

python -m aurarouter_antigravity --help

Testing & Verification

Run the full test suite:

pytest C:\projects\auracore\aurarouter-antigravity\tests -v

Run live vendor canary verification (requires GEMINI_API_KEY):

pytest C:\projects\auracore\aurarouter-antigravity\tests\test_live_vendor.py -v -m live_vendor

License

Apache-2.0