Google Antigravity SDK adds local model support for token efficiency and code privacy

Google has added local model support to the Antigravity SDK, letting developers run agentic workflows on their own machines while mixing in cloud models for higher-level tasks. In a demo auditing three Python modules, 97.2% of tokens ran locally and no source code was uploaded to the cloud.

• Hybrid orchestration: architect in the cloud, builders run locally
• In the demo, 97.2% of ~3400 tokens were processed locally
• Initial model is Gemma 4 26B A4B via Google's LiteRT
• Supports Ollama, LM Studio and vLLM via LocalOpenAIAgentConfig