What happened
Google has added local-model support to the Antigravity SDK, enabling developers to build and run agentic workflows on their own machines. When needed, those models can also be paired with cloud options for more demanding tasks.
The company says this hybrid setup offers more control over data privacy, inference costs, network dependencies and how computing load is shared. Google also outlines a model in which a main agent sets the plan, while several specialist agents handle smaller tasks in isolated environments.
In the demonstration cited, Google says most tokens were processed locally and no source code was sent to the cloud. The first implementation centres on Gemma 4 26B A4B via LiteRT, but the SDK also supports OpenAI-compatible local servers, including Ollama, LM Studio and vLLM.
Key points
- The Antigravity SDK now supports local models.
- Agentic workflows can run on developers’ own machines.
- Google still allows local and cloud models to be combined.
- The company says most of the work was done locally.
- The SDK also supports OpenAI-compatible local servers.
Why it matters for your organisation
For companies and technical teams, combining local execution with cloud options can reduce external dependencies and give more control over data and costs. It may also make it easier to keep sensitive parts of the workflow off the cloud while still using more complex agents when useful.