# Smart Connections - Chat Interface Workflow ## What This Feature Does The Smart Chat interface lets you have AI-powered conversations using your vault as the knowledge base. Unlike generic AI chat, this chat can *only* reference the notes you explicitly include in the conversation context—giving you precise control over what information the AI can access. ## Pre-Chat Checklist ✅ Smart Connections plugin installed and embeddings generated ✅ Understand which notes are relevant to your question ✅ Know how to build context (covered below) ✅ System prompt configured (if customizing AI behavior) ## Opening the Chat Interface **Method 1: Command Palette** 1. Press **⌘ P** (Cmd+P) 2. Type "Smart Chat" 3. Select "Smart Connections: Open Smart Chat" **Method 2: Sidebar** 1. Open Smart Connections sidebar 2. Click chat icon at top **Method 3: Ribbon Icon** (if enabled) - Click Smart Chat icon in left ribbon ## Chat Interface Components ### 1. History & Names (Top Bar) **Purpose:** Manage multiple conversation threads **Features:** - **Chat list** - View all past conversations - **Rename chat** - Click name to rename current conversation - **New chat** - Start fresh conversation with new context - **Revisit conversations** - Return to any past chat thread **Workflow:** ``` Research session on PKM methods → Rename to "PKM Research Nov 2025" Discussion about project planning → Rename to "Q4 Planning Ideas" ``` This lets you maintain organized, topic-specific conversation histories. ### 2. System Prompt (Bottom Left) **Purpose:** Set the AI's role, style, or constraints before sending messages **How to use:** 1. Click "Edit system message" at bottom of chat 2. Type your desired AI behavior 3. Close editor (prompt saved automatically) **Example prompts:** ``` "You are a personal knowledge management expert. Give concise, actionable advice based on the notes provided." "Act as a research assistant. Synthesize information from the provided notes and identify gaps or contradictions." "You're a writing coach. Help me connect ideas from my notes into a cohesive narrative structure." ``` **Pro tip:** Set system prompt *before* your first message for consistent behavior throughout the conversation. ### 3. Context Builder (Core Feature) **Purpose:** Select exactly which notes the AI can reference **Opening the builder:** - Click "Edit context" button - Type to search suggestions - Or use @ mentions in chat input #### Building Context: Step-by-Step **Step 1: Add Notes Manually** 1. Click "Edit context" 2. Type to search your vault 3. Suggestions appear as you type 4. Click note to add to context **Step 2: Expand with Connections** - Click "Show Connections" on any context item - Semantically related notes added automatically - Based on vector similarity **Step 3: Include Linked Notes** - Click "Show Links" on any context item - Notes with `[[wikilinks]]` to/from that note added - Follows your manual link structure **Step 4: Review and Refine** - See all context items listed - Click **X** next to any item to remove - Click "New" button to clear all context and start over #### Context Builder Example Workflow **Scenario:** You want to ask about improving your daily review habit ``` 1. Type "daily review" in context search → Add: "Daily Review Habit.md" 2. Click "Show Connections" → Auto-adds: "Weekly Planning.md", "Morning Routine.md", "Reflection Practice.md" 3. Click "Show Links" on "Daily Review Habit.md" → Adds: "Atomic Habits Notes.md", "Goal Tracking.md" 4. Review context: 6 notes total → Remove "Morning Routine.md" (not relevant) → Final context: 5 notes 5. Ask: "Based on my notes, what's missing from my daily review practice?" ``` ### 4. Using @ Mentions **Quick context addition without opening builder** **Syntax:** ``` Based on @[[Daily Review Habit]] and @[[Goal Tracking]], how can I connect these practices? ``` **Behavior:** - Type `@` in chat input - Start typing note name - Autocomplete suggestions appear - Select note to add to context for this message only **When to use:** - Quick questions referencing 1-2 notes - Don't need broader context set - One-off queries ## Complete Chat Workflow ### Basic Workflow **Step 1: Build Context** 1. Click "Edit context" 2. Add 3-5 core notes related to your question 3. Use "Show Connections" to expand (optional) 4. Review context list **Step 2: Set System Prompt** (Optional) 1. Click "Edit system message" 2. Define AI's role/behavior 3. Close editor **Step 3: Ask Question** 1. Type your question in chat input 2. Press Enter or click send 3. AI responds using *only* the context notes you provided **Step 4: Follow-Up** 1. Refine context if needed (add/remove notes) 2. Ask follow-up questions 3. Context persists throughout conversation **Step 5: Save Conversation** 1. Rename chat with meaningful title 2. Conversation saved in history 3. Revisit anytime with same context ### Advanced Workflow: Research Synthesis **Scenario:** Synthesize ideas across 20+ notes on a broad topic **Step 1: Start with Core Concepts** ``` Add: "Second Brain Overview.md" Add: "PKM Principles.md" Add: "Zettelkasten Method.md" ``` **Step 2: Expand Automatically** ``` Click "Show Connections" on each → 15 more notes added ``` **Step 3: Refine Context** ``` Remove: Off-topic notes Keep: 18 highly relevant notes ``` **Step 4: Ask Synthesis Questions** ``` "What are the common themes across these PKM approaches?" "Where do these methods disagree?" "What synthesis emerges from combining these ideas?" ``` **Step 5: Iterate** ``` Based on response, add more specific notes Remove broad overview notes Ask deeper questions ``` ## When to Use Smart Chat ### Use Smart Chat For: - **Synthesizing information** across multiple notes - **Identifying patterns** in your thinking over time - **Getting writing help** using your research as source material - **Question exploration** - "What do I think about X based on my notes?" - **Gap analysis** - "What am I missing in this topic area?" ### Use Traditional Tools For: - **Exact quotes** - Use Obsidian search - **Navigation** - Use graph view or backlinks - **Creating structure** - Use outline mode - **Quick reference** - Use Smart Connections sidebar (non-chat) ## Privacy and Control ### What the AI Can Access - **ONLY** notes you explicitly add to context - **ONLY** during the active chat session - **NO** access to other notes in your vault - **NO** internet connection required ### What the AI Cannot Access - Notes not in current context - Other chat conversations - Your file system outside the vault - Any external data sources **This means:** - You have precise control over information exposure - Safe to work with sensitive notes (just don't add them to context) - Great for working with client data, personal notes, etc. ## Pro Tips ### Tip 1: Build Reusable Context Sets For recurring questions, build context once and save the chat: ``` "Weekly Planning" chat → Always includes: Goals, Projects, Calendar notes "Writing Research" chat → Includes: Writing process notes, examples, quotes "Client X" chat → All notes tagged with #ClientX ``` ### Tip 2: Use System Prompts to Shape Output ``` For brainstorming: "Be creative and suggest novel connections" For analysis: "Be critical and identify weaknesses" For synthesis: "Combine ideas into coherent frameworks" ``` ### Tip 3: Layer Context Progressively Start narrow, expand as needed: ``` 1. Start with 2-3 core notes 2. Ask initial question 3. If response lacks depth, add more context 4. Ask follow-up with enriched context ``` ### Tip 4: Export Chat Results to Notes After valuable synthesis: 1. Copy key insights from chat response 2. Create new permanent note 3. Link to source notes used in context 4. Tag appropriately for future discovery ## Keyboard Shortcuts | Action | Shortcut | |--------|----------| | Send message | Enter | | Add to context | ⌘/Ctrl + Enter | | Open named context in builder | ⇧⌘/Ctrl + Enter | | Close context builder | Esc | | New line in message | ⇧ Enter | ## Troubleshooting ### Chat gives generic responses **Problem:** AI not using your notes **Solution:** - Verify notes are actually added to context (check context list) - Ensure notes have relevant content (not just titles) - Try more specific questions ### Context seems incomplete **Problem:** Missing relevant notes **Solution:** - Use "Show Connections" more liberally - Check if notes need better semantic content (add more descriptive text) - Manually search and add specific notes you know exist ### Responses feel inconsistent **Problem:** System prompt undefined or unclear **Solution:** - Set clear system prompt defining AI's role - Keep system prompt consistent throughout conversation - Start new chat if you want different AI behavior ## See Also - [[Smart Connections - Overview]] - Main plugin documentation - [[Vector Embeddings for Note Retrieval]] - How semantic search works - [[Building a Second Brain]] - Context for using chat effectively - [[Synthesis of Information]] - Creating new insights from existing notes