Gemma 4 12B on Laptop: Local AI Workflows

Key points

• 1

  Local Coding with Google AI Edge Gallery: Google AI Edge Gallery (macOS) now runs Gemma 4 12B to generate and execute Python code directly on your laptop. For example, you can describe data analysis goals in natural language, and the app will dynamically create, run, and visualize charts from text files—like comparing top girl names born in 2024 vs 2025. This eliminates cloud dependency for real-time data processing, letting you work offline while maintaining accuracy. To try it: download the app from Google’s site and use the 'Describe your goal' prompt to generate code.

• 2

  Voice-Driven Editing with Eloquent: Google AI Edge Eloquent (macOS) now uses Gemma 4 12B to polish text via voice commands entirely on your device. You can highlight text and say, 'restructure these notes into an executive summary' or 'translate this into Hindi'—with 60%+ better quality than prior models. This works without internet, using a customizable hotkey to dictate across apps. For immediate use: install the app, enable voice editing, and test with a paragraph of text to see real-time rewrites.

• 3

  LiteRT-LM CLI for Local Serving: The LiteRT-LM CLI now includes a 'serve' command to run Gemma 4 12B as an on-device API endpoint. You can import the model via `litert-lm import --from-huggingface-repo=litert-community/gemma-4-12B-it-litert-lm` and start a server with `litert-lm serve`. This lets you connect to tools like OpenWebUI for tasks such as generating particle effects without internet. To implement: run the CLI commands in your terminal, then use the endpoint via `curl http://localhost:9379/v1/chat/completions` for real-time interactions.