Google adds local model support to Antigravity SDK
The update keeps more code on-device while shifting only a small part of agent work to the cloud.
Google has added local-model support to the Antigravity SDK, letting developers run agent workflows on their own machines while still using cloud models for higher-level planning. The change is meant to reduce data exposure, lower inference costs, and cut reliance on networked cloud execution. In Google’s demo, it used about 95 of roughly 3,400 tokens in the cloud and kept the rest local, while also saying no source code left the device. The first supported path uses Gemma 4 26B A4B via LiteRT, but the SDK can also connect to OpenAI-compatible local servers such as Ollama, LM Studio, and vLLM.
Why it matters
For developers using Antigravity, the SDK can now run parts of an agent workflow locally instead of sending everything to cloud services. That matters because Google says the change is meant to improve privacy and cost control, and its demo kept source code off the cloud while using only a small share of tokens there. The first path runs through Gemma 4 26B A4B via LiteRT, with support also for local servers like Ollama, LM Studio and vLLM.
Keep or strike?
Does this story matter, or is it hype? Mark it before you see what everyone else did.
Sources
- InfoQ