fix minor bug embedding

CognioBugs.md

@@ -0,0 +1,459 @@
+# Cognio MCP - Bug Reports & Feature Requests
+
+**Date**: November 11, 2025  
+**Tested Version**: v1.0.8  
+**Tester**: Real-world usage with 15+ memories
+
+---
+
+## 🐛 CRITICAL BUGS
+
+### 1. Search Exact Match Failure
+**Severity**: HIGH  
+**Status**: ❌ Broken
+
+**Problem**:
+- Exact keyword search returns 0 results
+- IP addresses not found: `100.90.115.16` → 0 results
+- Technical specs not found: `B2als v2 4GB` → 0 results
+- Even though the data exists in memory
+
+**Test Results**:
+```
+✅ WORKS:
+- "XFCE desktop" → Found 2 (score 0.492, 0.427)
+- "Azure free tier" → Found 5 (score 0.757 - 0.507)
+- "Tailscale" → Found 2 (score 0.559, 0.523)
+
+❌ FAILED:
+- "100.90.115.16" (exact IP) → 0 results
+- "B2als v2 4GB" (exact specs) → 0 results
+- "rdpdemo VM 8GB B2ms" → 0 results
+```
+
+**Root Cause**:
+- Semantic search (all-MiniLM-L12-v2) is optimized for natural language
+- Poor performance on technical data (IPs, specs, numbers)
+- Similarity threshold (0.4) too high for exact matches
+
+**Suggested Fix**:
+1. Implement hybrid search (semantic + keyword/exact match)
+2. Add fallback to exact string matching if semantic returns 0
+3. Lower similarity threshold for technical content
+4. Consider better embedding model for technical data
+
+---
+
+### 2. ~~open_nodes Tool Always Fails~~ → FIXED: get_memory Tool Added
+**Severity**: HIGH  
+**Status**: ✅ FIXED
+
+**Problem**:
+- No way to retrieve full memory content by ID
+- `list_memories` only shows truncated snippets
+- Users couldn't read complete memory text
+
+**Solution Implemented**:
+1. **Added GET `/memory/{memory_id}` API endpoint** (src/main.py)
+   - Returns full memory content with all metadata
+   - Proper error handling for non-existent IDs (404)
+   - Includes project, tags, timestamps
+   
+2. **Added `get_memory` MCP tool** (mcp-server/index.js)
+   - Takes `memory_id` parameter
+   - Returns formatted memory details with full text
+   - Proper error handling and logging
+   
+3. **Added test coverage** (tests/test_api.py)
+   - Test for successful memory retrieval
+   - Test for 404 on non-existent memory
+   - ✅ All tests passing
+
+**Usage**:
+```javascript
+// MCP Tool
+mcp_cognio_get_memory({ memory_id: "abc-123-def" })
+
+// API Endpoint
+GET http://localhost:8080/memory/{memory_id}
+```
+
+**Files Modified**:
+- `src/main.py` - Added GET endpoint before DELETE endpoint
+- `mcp-server/index.js` - Added tool definition and handler
+- `mcp-server/README.md` - Added documentation
+- `tests/test_api.py` - Added test case
+
+**Test Results**:
+```bash
+$ pytest tests/test_api.py::test_get_memory -v
+✅ PASSED - Memory retrieval works correctly
+✅ PASSED - 404 error for non-existent memory
+```
+
+---
+
+### 3. list_projects Tool Error
+**Severity**: MEDIUM  
+**Status**: ✅ FIXED
+
+**Problem**:
+```javascript
+mcp_cognio_list_projects()
+// Returns: Error: Cannot convert undefined or null to object
+```
+
+**Expected Behavior**:
+- Should list all available projects
+- Help users discover existing projects
+
+**Suggested Fix**:
+- Handle null/undefined projects gracefully
+- Return empty array if no projects exist
+- Add proper error handling
+
+**Fix Implemented**:
+- Read from `stats.memories_by_project` with null-safe handling
+- Return JSON array `[{ "name": string, "count": number }]` or `[]` when empty
+- Improved error handling and predictable output format
+
+**Verification**:
+`mcp_cognio_list_projects()` → returns `[ {"name":"Cognio", "count": 1 } ]` after saving sample memory.
+
+---
+
+## ⚠️ LIMITATIONS
+
+### 4. list_memories Shows Only Snippets
+**Severity**: MEDIUM  
+**Status**: ✅ FIXED/ENHANCED
+
+**Problem**:
+- `list_memories` only shows preview/truncated text
+- No way to read full memory content via MCP tools
+- Users must export to file to read full content
+
+**Current Workaround**:
+- Use `export_memories` → read file manually
+- Not ideal for programmatic access
+
+**Suggested Enhancement**:
+- Add parameter `full_text: boolean` to `list_memories`
+- Or fix `open_nodes` to enable full text retrieval
+- Add `get_memory_by_id` tool for direct access
+
+**Fix Implemented**:
+- Added `full_text: boolean` to MCP `list_memories` (no truncation when true)
+- Added `get_memory` MCP tool to retrieve full memory by ID
+- Corrected paging (`page`/`limit`) and tags filtering in MCP
+
+**Usage**:
+```
+// Full text listing (MCP)
+mcp_cognio_list_memories({ project: "Cognio", page: 1, limit: 10, full_text: true })
+
+// Full memory by ID (MCP)
+mcp_cognio_get_memory({ memory_id: "<id>" })
+```
+
+**Verification**:
+- Listing shows full text when `full_text: true`
+- `get_memory` returns complete content and metadata
+
+---
+
+## 🚀 FEATURE REQUESTS
+
+### 5. Hybrid Search (Semantic + Keyword)
+**Priority**: HIGH
+
+**Description**:
+Implement dual search strategy:
+1. Try semantic search first
+2. If results < threshold, fallback to keyword/exact match
+3. Combine and rank results
+
+**Benefits**:
+- Best of both worlds
+- Handles natural language AND technical data
+- Better user experience
+
+**Example**:
+```javascript
+search_memory({
+  query: "100.90.115.16",
+  mode: "hybrid" // auto, semantic, keyword
+})
+```
+
+---
+
+### 6. Configurable Similarity Threshold
+**Priority**: MEDIUM
+
+**Description**:
+Allow users to adjust similarity threshold per search:
+
+```javascript
+search_memory({
+  query: "Azure VM",
+  similarity_threshold: 0.3 // default: 0.4
+})
+```
+
+**Benefits**:
+- More flexible search
+- Users can tune for their use case
+- Better for technical vs natural language content
+
+---
+
+### 7. Search Filters & Advanced Query
+**Priority**: MEDIUM
+
+**Description**:
+Add more search capabilities:
+
+```javascript
+search_memory({
+  query: "Azure",
+  tags: ["deployment", "current-setup"],
+  date_from: "2025-11-01",
+  date_to: "2025-11-11",
+  sort_by: "created_at" // score, created_at, updated_at
+})
+```
+
+**Benefits**:
+- More precise search results
+- Better organization
+- Time-based filtering
+
+---
+
+### 8. Bulk Operations
+**Priority**: LOW
+
+**Description**:
+Add tools for bulk operations:
+
+```javascript
+// Bulk update tags
+update_memory_tags({
+  memory_ids: ["id1", "id2", "id3"],
+  add_tags: ["new-tag"],
+  remove_tags: ["old-tag"]
+})
+
+// Bulk delete
+delete_memories({
+  memory_ids: ["id1", "id2", "id3"]
+})
+
+// Bulk move to project
+move_memories({
+  memory_ids: ["id1", "id2"],
+  target_project: "NewProject"
+})
+```
+
+**Benefits**:
+- Easier memory management
+- Save time on repetitive tasks
+
+---
+
+### 9. Memory Versioning
+**Priority**: LOW
+
+**Description**:
+Track memory changes over time:
+
+```javascript
+get_memory_history({
+  memory_id: "abc123"
+})
+// Returns: [v1, v2, v3] with timestamps
+
+restore_memory_version({
+  memory_id: "abc123",
+  version: 2
+})
+```
+
+**Benefits**:
+- Undo accidental changes
+- Track knowledge evolution
+- Better for collaborative use
+
+---
+
+### 10. Import Memories
+**Priority**: MEDIUM
+
+**Description**:
+Currently only export works, add import:
+
+```javascript
+import_memories({
+  file_path: "backup.json",
+  project: "Azure-Expert-Knowledge",
+  merge_strategy: "skip_duplicates" // overwrite, skip_duplicates, merge
+})
+```
+
+**Benefits**:
+- Backup & restore
+- Share memories between instances
+- Migration between servers
+
+---
+
+## 📊 PERFORMANCE SUGGESTIONS
+
+### 11. Caching for Frequent Searches
+**Priority**: LOW
+
+**Description**:
+- Cache search results for common queries
+- Invalidate on memory updates
+- Configurable TTL
+
+**Benefits**:
+- Faster repeated searches
+- Reduced embedding computation
+- Better UX
+
+---
+
+### 12. Batch Embedding Generation
+**Priority**: LOW
+
+**Description**:
+- Generate embeddings in batches during save
+- Async processing for large texts
+- Progress indicator
+
+**Benefits**:
+- Faster save operations
+- Better for bulk imports
+- Non-blocking
+
+---
+
+## 🔧 DEVELOPER EXPERIENCE
+
+### 13. Better Error Messages
+**Priority**: MEDIUM
+
+**Description**:
+Current errors are vague:
+- "Tool execution failed" → Why?
+- "Cannot convert undefined or null to object" → Where?
+
+**Suggested**:
+- Add detailed error messages
+- Include context (which field, what value)
+- Suggest fixes
+
+---
+
+### 14. Debug Mode
+**Priority**: LOW
+
+**Description**:
+Add debug flag to see internals:
+
+```javascript
+search_memory({
+  query: "Azure",
+  debug: true
+})
+// Returns: {
+//   results: [...],
+//   debug: {
+//     embedding_time: "50ms",
+//     search_time: "120ms",
+//     similarity_scores: [0.8, 0.6, 0.4],
+//     threshold_used: 0.4
+//   }
+// }
+```
+
+**Benefits**:
+- Easier troubleshooting
+- Performance optimization
+- Better understanding of search behavior
+
+---
+
+## 📝 DOCUMENTATION REQUESTS
+
+### 15. Search Best Practices
+**Priority**: MEDIUM
+
+**Topics Needed**:
+- When to use semantic vs keyword search
+- How to structure memory text for better search
+- Optimal similarity threshold for different use cases
+- Tag strategy recommendations
+
+---
+
+### 16. API Examples
+**Priority**: MEDIUM
+
+**Missing Examples**:
+- Complex search queries
+- Memory organization patterns
+- Project management workflows
+- Integration with other MCP servers
+
+---
+
+## 🎯 SUMMARY
+
+**Critical Issues** (Fix ASAP):
+1. ❌ Search exact match failure
+2. ❌ open_nodes tool broken
+3. ❌ list_projects error
+
+**High Priority Features**:
+1. 🚀 Hybrid search (semantic + keyword)
+2. 🚀 Full text retrieval (fix open_nodes or add get_memory_by_id)
+3. 🚀 Import memories
+
+**Nice to Have**:
+- Configurable similarity threshold
+- Advanced search filters
+- Bulk operations
+- Memory versioning
+- Better error messages
+
+---
+
+## 💡 TESTING NOTES
+
+**Test Environment**:
+- 15 memories in Azure-Expert-Knowledge project
+- Mix of technical (IPs, specs) and natural language content
+- Real-world usage scenario (Azure infrastructure documentation)
+
+**What Works Well**:
+- ✅ save_memory (reliable)
+- ✅ delete_memory (works)
+- ✅ export_memories (JSON & markdown)
+- ✅ set/get active project
+- ✅ Semantic search for natural language
+
+**What Needs Work**:
+- ❌ Exact match search
+- ❌ open_nodes tool
+- ❌ list_projects tool
+- ⚠️ Full text retrieval
+
+---
+
+**Repo**: https://github.com/0xrelogic/cognio-mcp  
+**Stars**: 45+ (but no issues reported yet!)  
+**Potential**: HIGH (great concept, needs polish)

src/embeddings.py

@@ -115,6 +115,9 @@ class EmbeddingService:
         vec1 = np.array(embedding1)
         vec2 = np.array(embedding2)
 
+        if vec1.shape != vec2.shape:
+            return 0.0
+
         dot_product = np.dot(vec1, vec2)
         norm1 = np.linalg.norm(vec1)
         norm2 = np.linalg.norm(vec2)