v1.3.0 "Deep Search" 🔍

What's New

🔍 Semantic Search

• QMD Integration — BM25 full-text search + vector similarity search, auto-detected at startup
• Embedding API Support — Configure any OpenAI-compatible embedding API (SiliconFlow, DeepSeek, Ollama, etc.)
• Three Search Modes — Text (exact match), BM25 (full-text ranking), Semantic (vector similarity)
• Smart Mode Toggle — Search modes only appear when the backend supports them
• Skeleton Loading — Smooth loading animation for semantic search results

⚙️ Settings Page

• New Settings UI — Configure embedding API (URL, key, model) with test connection button
• Embedding Cache Stats — View cached files count, coverage %, cache size, and active model
• SQLite Persistence — Embedding vectors cached to disk, survive server restarts

📅 Timeline Rewrite

• Single-request Loading — Replaced N+1 file fetches with one /api/timeline endpoint (38 requests → 1)
• Tailwind Styling — Consistent with the rest of the app
• Tag Filtering — Click tags to filter diary entries
• Daily Notes Removed — Timeline is now the single entry point for diary browsing

🎨 UI Improvements

• Phosphor Icons — Migrated from Lucide to Phosphor Icons (6000+ icons, weight variants)
• Sidebar Fixes — Active state no longer highlights multiple items simultaneously
• Markdown Themes — Multiple preview themes (medium, classic-typo)
• Agent Status Page — View config, gateway, and heartbeat status

⚡ Performance

• Timeline API — Server-side diary metadata extraction, instant page load
• Embedding Cache — SQLite-backed, only re-embeds when files change
• QMD Auto-detect — No overhead when QMD is not installed

🐛 Fixes

• Sidebar active state for Today/Timeline/Dashboard
• Dark mode theme support
• Hash route persistence for all views
• Shiki dual theme follows dark/light mode

---

Full changelog: v1.2.0...v1.3.0

v1.2.0 "Road Trip" 🚗

What's New

✨ Features

• PWA Support — Install as standalone app on mobile, desktop, and Tesla. Works offline.
• Hash-based Routing — URLs like #/file/memory/2026-01-31.md let you bookmark and share specific files. Browser back/forward works.
• Mermaid Diagrams — Fenced mermaid code blocks render as beautiful SVG diagrams with theme-aware styling.
• Auto-refresh Polling — Files auto-poll every 10s as WebSocket fallback, plus a manual refresh button.

⚡ Performance

• Express → Hono — Replaced Express stack with Hono (~14KB). Faster, lighter, better TypeScript support.
• CodeMirror Lazy Loading — First-paint gzip: 535KB → 310KB.
• Vendor Chunk Splitting — Separate chunks for react, codemirror, markdown, icons.
• Tesla/Large Screen Optimization — 24px base font, 56px touch targets, wider sidebar.

🐛 Fixes

• Code blocks without language tags now render properly.
• Monospace font enforced in <pre> blocks for correct CJK/ASCII art alignment.

---

Full changelog: v1.1.0...v1.2.0

v1.1.0 — Conflict Detection & Changelog

What's New

🔒 Optimistic Locking for Concurrent Edits
When saving a file that was modified on disk while you were editing, a conflict dialog appears with three options:

• Overwrite — force save your version
• Reload — discard edits, load the latest version
• Cancel — keep editing

This prevents data loss when the agent writes to a file during human editing.

📋 In-App Changelog
Click the version number in the sidebar footer to view the full changelog within the app.

v1.0.0 — Initial Release

🎉 First release of Memory Viewer

A web UI for browsing and editing AI agent memory files.

Features:

• 📁 File tree sidebar with collapsible directories
• 📖 GitHub-flavored Markdown rendering with syntax highlighting
• ✏️ In-browser editing with Ctrl+S save
• 🔍 Full-text search across all files (Ctrl+K)
• 📊 System dashboard — uptime, memory, load, today's summary
• 🔄 Live reload via WebSocket
• 🌗 Dark/light theme toggle
• 🔒 Sensitive content masking
• 🌐 i18n support (English & Chinese)
• 🤖 Multi-bot remote connections via Gateway API
• 📱 Responsive mobile layout