Google updates its Antigravity SDK to let developers run offline, agentic workflows locally. The update adds support for local execution using models such as Gemma 4 (26B A4B) with Google AI Edge LiteRT, enabling agents to perform tasks directly on-device.
The update also targets hybrid execution patterns. According to Google, developers can configure workflows where a cloud model acts as a lightweight planner, while local agents handle more token-intensive steps such as code auditing and patching on-device. This approach is presented as a way to balance planning capabilities with privacy and resource constraints.
Google further says the SDK offers “drop-in” compatibility with OpenAI-compatible inference servers, naming examples including Ollama and vLLM. Tech outlets describe the same focus on hybrid orchestration, where local models can be paired with cloud services rather than replacing them entirely. Overall, the update emphasizes privacy-first local utilities and flexible agent deployment, from fully offline to cloud-plus-local hybrid setups.
