Optimize model loading and improve the user experience.

Optimize model loading and improve the user experience.

This update brings a smarter and more reliable offline knowledge experience: • Improved local search with stronger coarse retrieval, semantic ranking, and neural reranking • More relevant answers through better source filtering and context selection • A more capable Note Agent that can create, read, append, rewrite, and delete notes using natural language • Safer note editing with previews and confirmation before changes are applied • Improved on-device speech recognition for more accurate and stable transcription • Replay voice recordings directly from notes and conversations • Export transcriptions for sharing, backup, or further editing • New OCR support for extracting text from images and PDFs • Fast, balanced, and high-quality OCR modes with progress tracking and cancellation • Smoother local AI through improved model scheduling, cancellation, memory handling, and session recovery • Refined Markdown editing, document import, chat, settings, and model management AI processing stays on-device for supported local models.

What’s New - Improved local AI conversations and document retrieval for more accurate, context-aware answers. - Enhanced Agent workflows with clearer progress updates, source references, and follow-up support. - Improved long-document summarization and cross-session content search. - Optimized local model loading, memory management, and device resource usage. - Improved Ask streaming performance to reduce lag and make conversations smoother. - Enhanced session switching, history loading, and overall stability.

This update improves the overall experience and adds support for more languages. - Improved overall app experience - Added multilingual support for more global users - Refined notes, documents, and voice workflows - Performance and stability improvements

This update focuses on making local AI faster, more stable, and easier to use. - Improved local model download, loading, switching, and memory handling for a smoother on-device AI experience - Added memory fit indicators for models, making it easier to choose the right model for your device - Improved Ask with better multi-turn conversations and more reliable local retrieval - Refined Settings, Notes, and first-run flows for a cleaner and smoother experience Based on current real-device testing: - 6GB memory is suitable for 2B-4B models - 8GB memory is suitable for models below 8B - 12GB memory can run a 9B model - 24GB memory can run a 26B model - 32GB memory can run 26B-31B models Actual results may still vary based on system memory usage, quantization, and context length.