# rohanpoudel2/solmari, Release Stories All notable changes to this project. ## Private second brain — text tools + on-device notes with semantic search _auto-2026-06-09 · 2026-06-09_ Solmari adds Notes feature with text tools and semantic search capabilities. ### What changed Introduces a capture-transform-store-retrieve surface for managing personal notes alongside chat. Features on-device semantic search, text transformation tools, and offline-first storage. Access via the drawer under Notes/Tools. ### Why it matters Solmari adds Notes feature with text tools and semantic search capabilities. ### New Features - Add Notes — private second brain with text tools and semantic search (#7), Introduces a capture-transform-store-retrieve surface for managing personal notes alongside chat. Features on-device semantic search, text transformation tools, and offline-first storage. Access via the drawer under Notes/Tools. ## Catalog-driven model selection with per-model config _auto-2026-06-07 · 2026-06-07_ Adds catalog-driven model selection to replace hardcoded engine pairs ### What changed Models are now selected from a recommended catalog with intelligent defaults per device. Users can download, switch between, and delete models with per-model configuration for prompts, sampling parameters, and context windows. ### Why it matters Adds catalog-driven model selection to replace hardcoded engine pairs ### New Features - Replace hardcoded engine pair with catalog-driven model selection (#6), Models are now selected from a recommended catalog with intelligent defaults per device. Users can download, switch between, and delete models with per-model configuration for prompts, sampling parameters, and context windows. ## Trim to core on-device chat, keyboard dictation, chat UX polish _auto-2026-06-06 · 2026-06-06_ Streamlines chat experience, replaces dictation with OS voice typing, polishes UI. ### What changed Removed non-essential features to focus on the foundational on-device AI chat functionality, reducing complexity and dependencies. Integrated native OS keyboard dictation in place of the broken custom dictation implementation, providing reliable cross-device voice input without additional dependencies. Enhanced the chat surface with UX improvements for a more polished messaging experience. ### Why it matters Streamlines chat experience, replaces dictation with OS voice typing, polishes UI. ### Breaking Changes - Streamline app to core on-device chat experience (#5), Removed non-essential features to focus on the foundational on-device AI chat functionality, reducing complexity and dependencies. ### New Features - Polish chat UI and interactions (#5), Enhanced the chat surface with UX improvements for a more polished messaging experience. ### Bug Fixes - Replace custom dictation with OS keyboard voice typing (#5), Integrated native OS keyboard dictation in place of the broken custom dictation implementation, providing reliable cross-device voice input without additional dependencies. ## Unique on-device features + dual inference backends (Apple FM + Gemma 4) _auto-2026-06-03-3 · 2026-06-03_ Adds persona presets, on-device semantic search, and dual inference backend support. ### What changed Introduces one-tap persona modes (Tutor, Coder, Editor, Brainstorm) accessible from the empty state and header. Each preset merges a specialized system message to shape AI behavior. Enables full-text semantic search across chat history using on-device processing, allowing users to quickly find relevant conversations. Extends the offline chat app to support both Apple Foundation Models and Gemma 4 as inference backends, giving users flexibility in model selection. ### Why it matters Adds persona presets, on-device semantic search, and dual inference backend support. ### New Features - Add persona presets for quick conversation modes (#4), Introduces one-tap persona modes (Tutor, Coder, Editor, Brainstorm) accessible from the empty state and header. Each preset merges a specialized system message to shape AI behavior. - Implement semantic chat search on-device (#4), Enables full-text semantic search across chat history using on-device processing, allowing users to quickly find relevant conversations. - Add dual inference backend support (#4), Extends the offline chat app to support both Apple Foundation Models and Gemma 4 as inference backends, giving users flexibility in model selection. ## Extrass _auto-2026-06-03-2 · 2026-06-03_ Adds AI chat interface with personas, document management, voice input, and embeddings for RAG. ### What changed Implemented core chat UI components including ChatScreen, ChatInputBar, ConversationDrawer, and EmptyState for managing conversations and user interactions. Introduced PersonaPicker component and personas library to enable users to select and customize AI assistant behaviors. Added DocumentChip component and RAG document retrieval system to support document-based context in conversations. Integrated speech-to-text functionality through use-dictation hook and STT service for hands-free chat input. Added embeddings service, vector store, and search indexing to enable semantic search capabilities across documents. Implemented storage, context-builder, and chat-provider for persisting conversations and building contextual AI prompts. Introduced reusable icon-symbol UI component for consistent iconography across the app. Updated app.json, package.json, and lock file to support new features and resolve dependencies. ### Why it matters Adds AI chat interface with personas, document management, voice input, and embeddings for RAG. ### New Features - Add chat interface with conversation management (#3), Implemented core chat UI components including ChatScreen, ChatInputBar, ConversationDrawer, and EmptyState for managing conversations and user interactions. - Add persona system for customizable AI interactions (#3), Introduced PersonaPicker component and personas library to enable users to select and customize AI assistant behaviors. - Implement document management for RAG (#3), Added DocumentChip component and RAG document retrieval system to support document-based context in conversations. - Add voice input support with dictation (#3), Integrated speech-to-text functionality through use-dictation hook and STT service for hands-free chat input. - Implement embeddings and vector search (#3), Added embeddings service, vector store, and search indexing to enable semantic search capabilities across documents. - Add chat persistence and context management (#3), Implemented storage, context-builder, and chat-provider for persisting conversations and building contextual AI prompts. - Add icon symbol component (#3), Introduced reusable icon-symbol UI component for consistent iconography across the app. ### Maintenance - Update project configuration and dependencies (#3), Updated app.json, package.json, and lock file to support new features and resolve dependencies. ## Turn sandbox into offline AI chat app _auto-2026-06-03 · 2026-06-03_ Converts sandbox to offline AI chat application ### What changed Convert the Foundation Models developer sandbox into a fully functional on-device chat app with streaming responses, multi-turn conversations, conversation persistence, and markdown-rendered replies. Add streamChat(messages, {abortSignal}) method to foundation-model for streaming multi-turn chat support. ### Why it matters Converts sandbox to offline AI chat application ### New Features - Transform sandbox into offline AI chat application (#2), Convert the Foundation Models developer sandbox into a fully functional on-device chat app with streaming responses, multi-turn conversations, conversation persistence, and markdown-rendered replies. Add streamChat(messages, {abortSignal}) method to foundation-model for streaming multi-turn chat support. ## Test _auto-2026-05-01 · 2026-05-03_ README documentation updated for clarity and completeness. ### What changed Update README documentation ### Why it matters README documentation updated for clarity and completeness. ### Maintenance - Update README documentation (#1)