M.I.M.I.R - Multi-agent Intelligent Memory & Insight Repository

# Vector Search UI Settings Guide ## Overview The Mimir Portal now includes a **Vector Search Settings** modal that allows users to configure semantic search parameters directly from the UI. These settings control how the AI retrieves relevant context from your knowledge graph during chat conversations. ## Accessing the Settings 1. Click the **Settings** cog icon (⚙️) located in the input area, to the left of the paperclip attachment icon 2. The settings modal will open, displaying all available configuration options ## Settings Configuration ### 1. Enable Vector Search - **Type**: Checkbox (On/Off) - **Default**: Enabled - **Description**: Master switch to enable or disable vector search entirely. When disabled, the AI will not search your knowledge graph for context. ### 2. Max Results - **Type**: Number input (1-50) - **Default**: 10 - **Description**: Maximum number of results to retrieve from the knowledge graph. Higher values provide more context but may slow down responses. - **Recommendation**: - **5-10**: For focused, specific queries - **15-25**: For comprehensive research queries - **25-50**: For deep exploration of related concepts ### 3. Min Similarity - **Type**: Number input (0.00-1.00) - **Default**: 0.80 - **Description**: Minimum cosine similarity threshold for results (0 = no filter, 1 = exact match). Higher values return only highly relevant results. - **Recommendation**: - **0.3-0.4**: Broad, exploratory search - **0.5-0.6**: Balanced relevance - **0.7-0.9**: Strict, high-precision results (default: 0.8) ### 4. Graph Depth - **Type**: Number input (1-3) - **Default**: 1 - **Description**: Graph traversal depth for multi-hop search. Determines how many relationship "hops" to explore from initial matches. - **Recommendation**: - **1**: Direct matches only (fastest) - **2**: Related concepts (balanced) - **3**: Deep network exploration (comprehensive, slower) ### 5. Node Types - **Type**: Multi-select checkboxes - **Default**: `todo`, `memory`, `file`, `file_chunk` - **Available Types**: - `todo` - Individual tasks - `todoList` - Collections of tasks - `memory` - Stored memories and research - `file` - Indexed files - `file_chunk` - File content chunks - `function` - Code functions - `class` - Code classes - `module` - Code modules - `concept` - Abstract concepts - `person` - People/contacts - `project` - Projects - `custom` - Custom node types - **Description**: Select which types of nodes to search. Only checked types will be included in results. ## Persistence All settings are automatically saved to **browser localStorage** when you click "Save Settings". They will persist across sessions and be applied to all future chat requests. ## API Integration When vector search is enabled, the settings are sent with each chat request: ```json { "messages": [...], "model": "gpt-4.1", "enable_tools": true, "tool_parameters": { "vector_search_nodes": { "limit": 10, "min_similarity": 0.8, "depth": 1, "types": ["todo", "memory", "file", "file_chunk"] } } } ``` ## Best Practices ### For General Chat ``` Enabled: ✓ Max Results: 10 Min Similarity: 0.8 Depth: 1 Types: memory, file, file_chunk ``` ### For Research & Exploration ``` Enabled: ✓ Max Results: 25 Min Similarity: 0.4 Depth: 2 Types: memory, file, file_chunk, concept ``` ### For Code-Related Queries ``` Enabled: ✓ Max Results: 15 Min Similarity: 0.6 Depth: 1 Types: file, file_chunk, function, class, module ``` ### For Task Management ``` Enabled: ✓ Max Results: 20 Min Similarity: 0.7 Depth: 2 Types: todo, todoList, project ``` ### Disable for Pure LLM Responses ``` Enabled: ✗ (Other settings ignored when disabled) ``` ## Reset to Defaults Click the **"Reset to Defaults"** button in the modal to restore the default settings: - Enabled: True - Max Results: 10 - Min Similarity: 0.8 - Depth: 1 - Types: `todo`, `memory`, `file`, `file_chunk` ## Troubleshooting ### No Results Found - **Lower min_similarity**: Try 0.3-0.4 for broader matching - **Increase max results**: Try 20-30 to capture more potential matches - **Check node types**: Ensure the types you need are selected - **Increase depth**: Try depth 2 for related concepts ### Too Many Irrelevant Results - **Raise min_similarity**: Try 0.7-0.8 for stricter matching - **Decrease max results**: Limit to top 5-10 most relevant - **Reduce depth**: Use depth 1 for direct matches only - **Narrow node types**: Select only the specific types you need ### Slow Responses - **Reduce max results**: Lower to 5-10 - **Decrease depth**: Use depth 1 instead of 2 or 3 - **Limit node types**: Select fewer types to search ## Technical Details ### Implementation - **Frontend**: `frontend/src/pages/Portal.tsx` - **State**: React useState with localStorage persistence - **API**: Passed as `tool_parameters.vector_search_nodes` in `/v1/chat/completions` requests ### Storage Key ```javascript localStorage.getItem('mimir-vector-search-settings') ``` ### Data Structure ```typescript interface VectorSearchSettings { enabled: boolean; limit: number; minSimilarity: number; depth: number; types: string[]; } ``` ## Future Enhancements Potential future additions: - Preset profiles (e.g., "Research Mode", "Code Mode", "Task Mode") - Per-conversation settings override - Real-time result preview in modal - Search statistics (avg results, avg similarity) - Node type auto-detection based on query ## Related Documentation - [Model Selection Guide](./MODEL_SELECTION.md) - [Chat API Documentation](../architecture/CHAT_API.md) - [Knowledge Graph Overview](../architecture/KNOWLEDGE_GRAPH.md) - [Vector Search Tool](../../src/tools/vectorSearch.tools.ts) --- **Last Updated**: 2025-11-19 **Version**: 1.0.0