AI Overhaul, Background SMS Import & Tier Restructuring

AI

• Migrated from llama.cpp to Gemma models (Gemma 4 E2B, E4B) via flutter_gemma
• Added lightweight Qwen 0.6B model option (586MB) for smaller devices
• Replaced ONNX embeddings with EmbeddingGemma 300M (768-dim vectors)
• AI tool calling with human-in-the-loop: propose transactions, budgets, and savings goals
• Read tools: search transactions, spending by category, balances, budgets, goals
• Apple Foundation Models integration on iOS 26+ — zero model download required
• Streaming responses for all model backends
• AI setup wizard for guided model download and configuration

SMS Import

• Background import with foreground service (Android) and BGProcessingTask (iOS)
• Live progress banner across all authenticated screens
• AI-enhanced parsing: Gemma LLM handles messages that regex can’t match
• Pending transactions review screen before confirmation
• Smart filtering of OTPs, promotions, and spam before parsing

Tier Restructuring

• 8 features moved from Premium to Free: multi-currency, custom categories, budget rollover, tags, split transactions, natural language queries, PDF export, goal milestones
• Premium refocused on analytics depth: advanced reports, trends, forecasts, anomaly detection, net worth history, recurring detection, widgets, investment analytics
• Supporter tier reduced to $29.99 one-time (lifetime premium)

UX

• FAB menu redesigned from radial arc to floating grid card
• Multi-currency form hardcoding removed across all forms
• Fixed overflow issues across 17 widget and screen files
• Biometric flicker and dark mode text fixes