Google brings Gemma 4 12B to everyday laptops – full agentic AI workflows, entirely offline, on 16GB RAM
Category: On-Device AI
What it is
Google DeepMind’s Gemma 4 12B is now available for local deployment via the Google
AI Edge stack on standard 16GB RAM machines. It powers three new tools: AI Edge
Gallery on macOS (generates and executes Python scripts for data analysis entirely on
device), AI Edge Eloquent (voice dictation and interactive text editing with no cloud
dependency), and a new LiteRT-LM CLI serve command that creates a local OpenAI
compatible endpoint for agentic workflows. The combination delivers multimodal,
code-executing AI without sending data to any server.
Why it Matters for Enterprises
On-device frontier AI changes the calculus for data-sensitive industries. Teams in legal,
finance, healthcare, and defence can now run capable agentic models on a laptop with
zero data leaving the device - making compliance-friendly AI deployment achievable
without cloud infrastructure or vendor lock-in.
