Gemma 4 dropped in April 2026, and the on-device Flutter story is genuinely exciting: streaming inference, native function calling, thinking mode, multimodal input, all running locally, with zero API costs and no data leaving the device. The flutter_gemma package makes it surprisingly approachable. And then you actually build something. This roundtable gathers Flutter developers currently working with or evaluating Gemma 4 and the flutter_gemma package. I will begin our conversation by sharing practical insights from the recent launch of Lumina Journal, an AI-powered journaling application that operates entirely on-device, powered by Gemma 4 E2B. Key topics for discussion include: * Technical challenges, such as silent GPU fallback on mid-range Android devices. * Performance metrics for function-calling reliability and thinking-mode token budgets. * User experience hurdles, such as the significant 2.5 GB initial download requirement. While documentation provides a baseline, the actual implementation often reveals unexpected complexities. This session is a collaborative forum to exchange lessons learned, address unanswered questions, and define the current state and future trajectory of on-device AI within the Flutter ecosystem as of 2026.