|
1 | 1 | # ChatGPT MD Changelog |
2 | 2 |
|
| 3 | +## v3.0.0 (December 2025) - Privacy-First AI Tool Calling |
| 4 | + |
| 5 | +### 🎯 Major Features |
| 6 | + |
| 7 | +- **Privacy-First Tool Calling System**: AI assistants can now search your vault, read files, and search the web—with explicit human approval at every step |
| 8 | + - Three-layer approval architecture: approve execution → review results → approve sharing |
| 9 | + - No data reaches the LLM without your explicit consent |
| 10 | + - All tools disabled by default (opt-in feature) |
| 11 | +- **Vault Search Tool**: AI can discover relevant notes in your vault |
| 12 | + - Multi-word OR search (matches ANY query word) |
| 13 | + - Searches both filenames and file content |
| 14 | + - Interactive results review: you select which files to share |
| 15 | + - Automatically excludes current file from search results |
| 16 | +- **File Read Tool**: AI can request access to specific files |
| 17 | + - Granular file selection in approval modal |
| 18 | + - Full content reading with user consent |
| 19 | + - Batch reading support for multiple files |
| 20 | +- **Web Search Tool (Experimental)**: AI can search the web using privacy-focused Brave Search API |
| 21 | + - 1,000 free queries/month with Brave Search API |
| 22 | + - Review and filter web results before sharing |
| 23 | + - Custom search provider support for self-hosted solutions |
| 24 | + - Optional full page content fetching |
| 25 | + |
| 26 | +### 🔒 Privacy & Control |
| 27 | + |
| 28 | +- **Human-in-the-Loop Architecture**: Every tool call requires explicit user approval before execution and before data sharing |
| 29 | +- **Multi-Stage Approval Process**: |
| 30 | + 1. Tool execution approval: see what the AI wants to do |
| 31 | + 2. Results review: see what the tool found |
| 32 | + 3. Selective sharing: choose exactly which results to share with AI |
| 33 | +- **Data Minimization**: Share only the specific results you approve—nothing more |
| 34 | +- **Full Transparency**: All tool requests displayed with clear descriptions of what will happen |
| 35 | +- **Audit Trail**: Debug mode logs all tool calls and decisions for troubleshooting |
| 36 | +- **Local Storage**: All API keys stored locally in Obsidian settings |
| 37 | + |
| 38 | +### 🏗️ Technical Architecture |
| 39 | + |
| 40 | +- **New Service Layer**: |
| 41 | + - `ToolService`: Orchestrates tool calling with approval workflow |
| 42 | + - `ToolRegistry`: Manages available tools and their configurations |
| 43 | + - `ToolExecutor`: Executes approved tools with security checks |
| 44 | + - `VaultTools`: Vault-specific tool implementations |
| 45 | + - `WebSearchService`: Web search integration with Brave Search API |
| 46 | + - `StreamingHandler`: Refactored streaming response processor |
| 47 | + - `Logger`: Comprehensive debug logging utility |
| 48 | +- **Type-Safe Tool System**: Complete TypeScript interfaces for tools, approvals, and results |
| 49 | +- **AI Service Refactoring**: All 6 AI services updated to support tool calling |
| 50 | + - OpenAI, Anthropic, Gemini, OpenRouter, Ollama, LM Studio |
| 51 | + - Consistent tool interface across all providers |
| 52 | + - Improved streaming response handling |
| 53 | + - Better error handling and user feedback |
| 54 | + |
| 55 | +### 🎨 User Interface |
| 56 | + |
| 57 | +- **ToolApprovalModal**: Interactive modal for approving tool execution requests |
| 58 | + - Shows tool name, parameters, and description |
| 59 | + - File selection UI for file_read tool |
| 60 | + - Clear approve/cancel actions |
| 61 | +- **SearchResultsApprovalModal**: Review and filter vault search results |
| 62 | + - Preview of found files with metadata |
| 63 | + - Multi-select interface for granular control |
| 64 | + - Shows file paths, names, and match counts |
| 65 | +- **WebSearchApprovalModal**: Review and filter web search results |
| 66 | + - Displays titles, URLs, and snippets |
| 67 | + - Multi-select interface for result filtering |
| 68 | + - Preview of web content before sharing |
| 69 | + |
| 70 | +### ⚙️ Configuration & Settings |
| 71 | + |
| 72 | +- **New Settings Section**: "Tool Calling" with comprehensive configuration options |
| 73 | + - Enable/disable tool calling (disabled by default) |
| 74 | + - Enable/disable web search |
| 75 | + - Brave Search API key configuration |
| 76 | + - Custom search provider support |
| 77 | + - Maximum web search results (1-10) |
| 78 | + - Debug mode for detailed logging |
| 79 | +- **Frontmatter Support**: Per-note tool calling configuration (coming soon) |
| 80 | +- **Provider Flexibility**: Use Brave Search (default) or custom search endpoints |
| 81 | + |
| 82 | +### 📦 Under the Hood |
| 83 | + |
| 84 | +- **40 Files Changed**: 3,632 insertions, 2,176 deletions |
| 85 | +- **Major Refactoring**: AI service implementations streamlined for tool support |
| 86 | +- **Enhanced Streaming**: Better real-time response processing with tool call handling |
| 87 | +- **Improved Error Handling**: Comprehensive error messages for tool failures |
| 88 | +- **Performance Optimizations**: Efficient vault search with configurable limits |
| 89 | +- **Code Quality**: Comprehensive TypeScript types for all tool-related functionality |
| 90 | + |
| 91 | +### 🔧 Developer Experience |
| 92 | + |
| 93 | +- **Debug Mode**: Detailed console logging for troubleshooting tool operations |
| 94 | +- **Extensible Architecture**: Easy to add new tools in the future |
| 95 | +- **Clear Interfaces**: Well-documented TypeScript interfaces for tool development |
| 96 | +- **Service Locator Pattern**: Consistent dependency injection for tool services |
| 97 | + |
| 98 | +### 🚀 Future Roadmap |
| 99 | + |
| 100 | +Potential future enhancements: |
| 101 | + |
| 102 | +- Additional tools: calendar integration, task management, graph analysis |
| 103 | +- Configurable auto-approval rules for trusted tools |
| 104 | +- Tool usage analytics (local only, privacy-focused) |
| 105 | +- Batch file operations with granular control |
| 106 | +- Advanced search operators and filters |
| 107 | + |
| 108 | +### 📚 Knowledge Management Benefits |
| 109 | + |
| 110 | +- **Context-Aware AI**: AI can discover relevant information from your vault |
| 111 | +- **Research Augmentation**: Combine vault knowledge with web search results |
| 112 | +- **Note Discovery**: Find connections between notes you might have missed |
| 113 | +- **Enhanced Productivity**: AI with access to your knowledge base (with your permission) |
| 114 | +- **Privacy-Preserved**: Full control over what information AI sees |
| 115 | + |
| 116 | +### 🎓 Use Cases |
| 117 | + |
| 118 | +- **Research Assistant**: "Search my vault for notes about quantum computing and find recent papers on the topic" |
| 119 | +- **Knowledge Synthesis**: "Find all my meeting notes from Q3 and summarize the key decisions" |
| 120 | +- **Cross-Reference**: "Search for notes mentioning both machine learning and productivity" |
| 121 | +- **Web-Enhanced Writing**: "Search the web for the latest statistics on climate change and incorporate them" |
| 122 | +- **Information Retrieval**: "Find my notes on project X and tell me the current status" |
| 123 | + |
3 | 124 | ## v2.8.1-beta (August 2025) - CORS-Free Streaming & Network Improvements |
4 | 125 |
|
5 | 126 | ### 🚀 Major Network Improvements |
|
0 commit comments