Local Models in the Antigravity SDK

Developers can now run Antigravity SDK agents on their own machine with Gemma 4, offline and with no API fees.

Initial support is for Gemma 4 26B A4B, run on-device through Google AI Edge's LiteRT runtime.
The SDK can also connect to local model servers that use the OpenAI API format, such as Ollama, LM Studio, and vLLM.
A hybrid setup lets a cloud model like Gemini 3.8 Flash plan the work while a group of local Gemma 4 instances carries it out on the machine.
Local runs have no token costs or rate limits, and code and requests stay on the machine.
Google recommends at least 24 GB of GPU memory or unified memory. The Gemma 4 26B checkpoint is about a 16.8 GB download.
Available now in the Antigravity Python SDK.

Agent workflows over code or data that should not leave the machine.

Privacy rules, API costs, or rate limits are blocking an agent workflow you want to run.

Developers building on the Antigravity Python SDK who have a machine with a strong GPU.

Join the newsletter to get ahead with AI.

I share strategies, new modules, free guides, and tools to get you better at using AI in your work.