Google Antigravity SDK Adds Support for Local AI Models

Summary

Google says the Antigravity SDK now supports local workflows across a range of models and execution options, beginning with Gemma 4 26B A4B through Google AI Edge’s LiteRT. Developers can run agentic assistance fully offline, keeping code and requests on the local machine and avoiding API costs or rate limits. Google recommends a machine with more than 24GB of VRAM or unified memory, and provides Python setup instructions for installing the SDK and LiteRT-LM, downloading the model, and connecting it to an agent. The SDK also supports hybrid orchestration: in Google’s example, Gemini 3.8 Flash plans a security audit from filenames and task descriptions, while local Gemma instances reproduce vulnerabilities, propose fixes, critique patches, and run regression tests. The example used 95 cloud tokens for planning, while 3,322 tokens, or 97.2% of the total, ran locally and offline; the article says the resulting patches were verified and passed the tests. Another example has Gemma generate and test a Python command-line resource monitor using psutil and rich. For broader backend flexibility, LocalOpenAIAgentConfig can connect the same orchestration and tools to OpenAI-compatible servers such as Ollama, LM Studio, and vLLM.