VaultBuddy v1.0.5

🚀 New Features

• Two‑stage retrieval
  Build a lightweight index via Obsidian metadataCache (title/headings/tags/aliases/date), rank by relevance, then read full content only for top‑ranked files. Greatly improves performance.

• Enhanced tokenization & stemming
  Supports Chinese word segmentation (Intl.Segmenter) + English stemming, plus bilingual synonym expansion for smarter search.

• Time‑based queries
  Natural language time conditions like “last X days/months”, “yesterday”, “this week”, combined with time + keyword scoring.

• Searched notes bar
  After AI responds, a “Searched Notes” bar appears at the bottom. Expand to view all referenced notes, click to navigate.

• Conversation token usage display

  • Input/output token count shown below each AI reply, formatted with k/m units
  • Cumulative token count and message count displayed in conversation header

• UI improvements

  • Clear view when deleting the current conversation or switching to a new one
  • Temperature slider shows its value live (setDynamicTooltip)

🐛 Bug Fixes

• Fixed sources panel CSS causing expand/collapse malfunction
• Fixed undocumented eslint-disable-next-line comment in ESLint review

🔧 Technical Improvements

• Dependency upgrades: TypeScript → 5.8.3, esbuild → 0.28, gpt-tokenizer → 3.4, @typescript-eslint → 8
• Improved chunking: heading‑aware splitting, overlapping windows, minimum chunk size
• Improved scoring: TF weighting, position bias, length penalty
• Added usage field to ChatMessage type, persisted with conversation data
• Updated .gitignore (added reasonix.toml and .codegraph/)

📄 Documentation

• Updated README: features as a list, added “How It Works” (two‑stage retrieval), removed Development section

VaultBuddy v1.0.4

Full Changelog: 1.0.3...1.0.4

VaultBuddy v1.0.3

✨ Improvements

• Redesigned settings tab switcher — Now a segmented card-style control for a more modern look
• Consolidated layout — Merged "Default Model" dropdown and "Add Model" button into a single row
• Strict model editing — Changes only persist on explicit save with all fields validated

🔧 Code Quality

• Fixed all @typescript-eslint type-safety warnings:
  • Catch blocks use unknown + instanceof Error instead of implicit any
  • Defined explicit interfaces for loadData() and API responses
  • Declared window.moment via declare global, removed as any
  • Unified async event callbacks to void + .then() pattern
• Annotated fetch usage with reason (requestUrl lacks streaming support)

📁 Changed Files

• src/ui/settings.ts — Settings UI redesign, model edit validation
• src/ui/view.ts — Type safety fixes, tab switcher styles
• src/core/i18n.ts — window.moment type declaration
• src/services/aiService.ts — API response interfaces, error type guards
• src/services/storage.ts — StoredData interface
• styles.css — Segmented control styles
• manifest.json / package.json / versions.json — Version bumped to 1.0.3

VaultBuddy v1.0.2

✨ Improvements

• Redesigned settings tab switcher — Now a segmented card-style control for a more modern look
• Consolidated layout — Merged "Default Model" dropdown and "Add Model" button into a single row for a cleaner interface

🐛 Bug Fixes

• Fixed an issue where clearing a field while editing an existing model and clicking Save would persist empty values
• Added validation on save — all 5 fields (Name, Base URL, API Key, Model ID, Context Window) must be non‑empty before changes are applied

VaultTalk v1.0.1

A maintenance release focused on multi-turn conversations, smarter AI behavior, safer rendering, and technical cleanup.

✨ Improvements

• Multi-turn conversation support — AI now remembers the last 20 messages, enabling natural follow-up questions (e.g., "summarize it as a table" now correctly references previous context)
• Smarter AI responses — Optimized system prompt with structured response style: more concise, direct, and less repetitive
• Better clarification — AI will ask for clarification instead of guessing when questions are ambiguous
• Safe markdown rendering — Replaced innerHTML with Obsidian's native MarkdownRenderer for secure rendering
• Wiki-link navigation fix — Folder links are now displayed as plain text to prevent accidental note creation; file links navigate correctly using exact vault paths

🔧 Technical Changes

• Removed marked dependency, using Obsidian's built-in markdown renderer
• Replaced navigator API with Platform API for OS detection
• Replaced localStorage with Obsidian's loadLocalStorage / saveLocalStorage
• Moved inline styles to CSS classes for better theme compatibility
• Updated minAppVersion to 1.8.7
• Cleaned up unused context builder methods

VaultTalk v1.0.0

VaultTalk is an AI-powered chat assistant that lives in your Obsidian sidebar. It turns your entire vault into context, so you can ask questions and get answers based on your actual notes.

Key Features:

• 💬 Sidebar Chat — A persistent AI assistant panel right inside Obsidian. Ask anything, get answers grounded in your notes.
• 📚 Vault as Context — Automatically reads your vault, selects the most relevant passages, and sends them as context to the AI. Every answer cites its sources with [[wiki-links]].
• ⚡ Streaming Responses — See answers appear in real-time as the AI generates them.
• 🔐 Encrypted API Key Storage — API keys are encrypted locally using the Web Crypto API. Never stored in plain text.
• 🤖 Custom Models — Supports any cURL-compatible OpenAI API (DeepSeek, OpenAI, Groq, etc.). Add multiple models and switch freely.
• 📝 Conversation History — All conversations are automatically saved. Resume where you left off anytime.
• 🌍 i18n — Chinese and English interface, auto-detected from your Obsidian language setting.
• 📱 Mobile Support — Works on both desktop and mobile.支持** — 桌面端和移动端均可使用。