Gemma 4 Local Chat lands on Mac

This is less a flashy demo than a credible local-LLM UX package: the model matters, but the real product is the frictionless offline workflow.

• –MLX on Apple Silicon plus an M4 16GB benchmark of roughly 19 tok/s makes the experience sound practical, not just theoretical
• –Persistent system prompts and full conversation memory push it beyond a basic wrapper and toward a useful writing or roleplay assistant
• –The desktop shortcut is the right kind of polish for local AI, because setup friction is usually the thing that kills adoption
• –Full-history memory will hit context limits fast, so the token counter is doing real work here, not just decoration
• –MIT licensing and offline operation after setup make it an attractive privacy-first reference implementation for local AI apps