google-antigravity
Google Antigravity SDK for building AI agents
What it is and what it does
Google Antigravity SDK is a Python framework for building AI agents that run on Google's Gemini models. It abstracts away the complexity of managing the agentic loop—the iterative cycle of reasoning, tool calling, and response generation—so you can define agent behavior through system instructions, custom tools, and policies without managing connection lifecycle or state manually.
The SDK provides both high-level convenience (the `Agent` class with async context manager) and lower-level control (direct `Conversation` and `ConnectionStrategy` APIs). It supports multimodal input (images, videos, audio, documents), custom Python functions as tools, integration with Model Context Protocol (MCP) servers, streaming responses and internal reasoning, and policy-based behavior control. It works with both Gemini API and Gemini Enterprise Agent Platform (formerly Vertex AI) via Application Default Credentials.
Use it for:
- Build a code analysis agent that reads repository files and documents, then streams architectural recommendations.
- Create an interactive assistant that accepts user queries and streams real-time responses with visible reasoning steps.
- Wire custom Python functions as tools so an agent can query databases, call APIs, or execute system commands autonomously.
- Connect MCP servers to expose external capabilities (e.g., file system tools, web search) to the agent without custom integration code.
- Deploy a read-only agent for document analysis or Q&A over PDFs and images without enabling write capabilities.
Worth the install?
AI-flagged interpretation of the facts on this page — verify before relying
Builds and runs AI agents powered by Google's Gemini, handling the agentic loop, tool wiring, and stateful conversation management so you focus on agent behavior rather than infrastructure.
Yes, with conditions. The SDK is actively maintained (released 1 day ago), permissively licensed, and solves a real problem—abstracting agentic loop complexity. However, it is in Alpha status, depends on 7 runtime libraries, requires Python 3.10+, and mandates installation from PyPI due to compiled binaries. Install if you are building Gemini-powered agents and can tolerate early-stage API changes; avoid if you need production stability guarantees or prefer lightweight agent frameworks.
Install
google-antigravity on PyPI
pip
pip install google-antigravityuv
uv add google-antigravitypoetry
poetry add google-antigravityInstalling google-antigravity
Before you install
Medium install friction due to platform-specific compiled binaries included in wheels; must install from PyPI, not from source. Active maintenance with recent release (1 day old) and 3011 repository stars suggest ongoing support.
License in practice
Apache-2.0 permissive license allows commercial and private use with minimal restrictions; suitable for most production scenarios.
Quickstart
pip install google-antigravity
import asyncio
from google.antigravity import Agent, LocalAgentConfig
async def main():
config = LocalAgentConfig()
async with Agent(config) as agent:
response = await agent.chat("Hello!")
print(await response.text())
asyncio.run(main())
Requires Python 3.10+. Relies on a compiled runtime binary included only in PyPI wheels; cloning the repository alone is insufficient. Requires GEMINI_API_KEY environment variable or explicit api_key parameter for Gemini API access.
Verify before relying
- Whether the 7 runtime dependencies (absl-py, google-genai, mcp, pydantic, uvicorn, websockets, protobuf) are all required or if some are optional for specific features.
- Performance characteristics and latency overhead of the agentic loop abstraction layer.
- Compatibility and tested integration status with specific MCP server implementations.
Package facts
| License | Apache-2.0 (permissive) |
| Python support | supports the current Python release (>=3.10) |
| Install friction | medium — platform-specific wheel |
| Runtime dependencies | 7 — absl-py, google-genai, mcp, pydantic, uvicorn, websockets, protobuf |
| Maintenance | actively maintained — 1 days since the last release |
| Last repo commit | |
| First released | |
| Downloads | 386,789/month — #7,049 on PyPI (30-day window, as of 2026-08-14) |
| Known vulnerabilities | none known (OSV.dev, checked 2026-08-14) |
Evidence: google_antigravity-0.1.12-py3-none-macosx_11_0_arm64.whl; google_antigravity-0.1.12-py3-none-manylinux_2_17_aarch64.whl; google_antigravity-0.1.12-py3-none-manylinux_2_17_x86_64.whl; google_antigravity-0.1.12-py3-none-win_amd64.whl; google_antigravity-0.1.12-py3-none-win_arm64.whl
Tags
More Artificial Intelligence packages
LiteLLM provides a unified Python interface to…
permissive · top 100 on PyPI
huggingface-hubClient library and CLI tool for downloading,…
permissive · top 100 on PyPI
langchainLangChain provides a framework for building…
permissive · top 1,000 on PyPI
hf-xethf-xet provides chunk-based deduplication and…
permissive · top 1,000 on PyPI
tokenizersTokenizers converts raw text into token…
permissive · top 1,000 on PyPI
transformersTransformers provides a unified framework for…
permissive · top 1,000 on PyPI
google-cloud-aiplatformPython SDK for building, deploying, and…
permissive · top 1,000 on PyPI
vertexaiPython SDK for building, deploying, and…
permissive · top 5,000 on PyPI
google-agents-cliA CLI toolchain that equips coding agents…
permissive · top 15,000 on PyPI
livekit-plugins-googleIntegrates Google Cloud AI services (Gemini,…
permissive · top 5,000 on PyPI
google-genaiPython SDK for integrating Google's Gemini…
permissive · top 1,000 on PyPI
trinity-agentTrinity orchestrates multiple AI coding agents…
permissive · top 15,000 on PyPI
llama-index-llms-google-genaiIntegrates Google's Gemini models into…
permissive · top 15,000 on PyPI
llama-index-llms-vertexIntegrates Google Vertex AI's LLM capabilities…
permissive · top 15,000 on PyPI
cugaCUGA is an enterprise-grade generalist agent…
permissive · top 15,000 on PyPI
praisonaiagentsPraisonAI agents is a Python framework for…
unclear · top 15,000 on PyPI