superlocalmemory 2.7.2 → 2.7.4

package/CHANGELOG.md

@@ -16,6 +16,35 @@ SuperLocalMemory V2 - Intelligent local memory system for AI coding assistants.
 
 ---
 
+## [2.7.4] - 2026-02-16
+
+### Added
+- Per-profile learning — each profile has its own preferences and feedback
+- Thumbs up/down and pin feedback on memory cards
+- Learning data management in Settings (backup + reset)
+- "What We Learned" summary card in Learning tab
+
+### Improved
+- Smarter learning from your natural usage patterns
+- Recall results improve automatically over time
+- Privacy notice for all learning features
+- Learning and backup databases protected together
+- All dashboard tabs refresh on profile switch
+
+---
+
+## [2.7.3] - 2026-02-16
+
+### Improved
+- Enhanced trust scoring accuracy
+- Improved search result relevance across all access methods
+- Better error handling for optional components
+
+### Fixed
+- Corrected outdated performance references in documentation
+
+---
+
 ## [2.7.1] - 2026-02-16
 
 ### Added
@@ -270,7 +299,7 @@ SuperLocalMemory V2 represents a complete architectural rewrite with intelligent
 - **Profile Management** - Multi-profile support with isolated databases
 
 ### Performance
-- 3.3x faster search (45ms vs 150ms in V1)
+- Improved search performance over V1 (see Performance Benchmarks)
 - 60-96% storage reduction with compression
 
 ### Research Foundation

package/README.md

@@ -425,7 +425,7 @@ WAL mode + serialized write queue = zero "database is locked" errors, ever.
 
 ### Storage
 
-10,000 memories = **13.6 MB** on disk (~1.9 KB per memory). Your entire AI memory history takes less space than a photo.
+10,000 memories = **13.6 MB** on disk (~1.4 KB per memory). Your entire AI memory history takes less space than a photo.
 
 ### Graph Construction
 

package/docs/ARCHITECTURE.md

@@ -381,7 +381,7 @@ SuperLocalMemory V2 uses a hierarchical, additive architecture where each layer
    - Cold storage archival
 
 **Performance:**
-- Full-text search: ~45ms (avg)
+- Full-text search: Sub-11ms median for typical databases (see wiki Performance Benchmarks for measured data)
 - Insert: <10ms
 - Tag search: ~30ms
 
@@ -737,7 +737,7 @@ Vue: 10% confidence → Low priority, exploratory
 | Operation | Complexity | Typical Time |
 |-----------|-----------|--------------|
 | Add memory | O(1) | <10ms |
-| Search (FTS5) | O(log n) | ~45ms |
+| Search (FTS5) | O(log n) | ~11ms median (100 memories) |
 | Graph build | O(n²) worst, O(n log n) avg | ~2s (100 memories) |
 | Pattern update | O(n) | <2s (100 memories) |
 | Find related | O(1) | <10ms |
@@ -794,12 +794,12 @@ Vue: 10% confidence → Low priority, exploratory
 
 ### Current Limits (Tested)
 
-| Memories | Build Time | Search Time | Database Size |
-|----------|-----------|-------------|---------------|
-| 20 | <0.03s | ~30ms | ~30KB |
-| 100 | ~2s | ~45ms | ~150KB |
-| 500 | ~15s | ~60ms | ~700KB |
-| 1000 | ~45s | ~100ms | ~1.5MB |
+| Memories | Build Time | Search Time (median) | Database Size |
+|----------|-----------|----------------------|---------------|
+| 100 | 0.28s | 10.6ms | ~150KB |
+| 500 | ~5s | 65.2ms | ~700KB |
+| 1,000 | 10.6s | 124.3ms | 1.50 MB |
+| 5,000 | 277s | 1,172ms | ~6.8 MB |
 
 ### Scaling Strategies
 

package/docs/COMPRESSION-README.md

@@ -245,7 +245,7 @@ Priority order:
 - Tier 2 (40 memories @ 10KB): 400KB
 - Tier 3 (30 memories @ 2KB): 60KB
 - **Total: 1.96MB (61% reduction)**
-- **Search time: 45ms** (only scan Tier 1+2)
+- **Search time: Sub-11ms median for typical databases** (only scan Tier 1+2, see wiki Performance Benchmarks)
 - **Memory load: 1.9MB** (Tier 3 loaded on-demand)
 
 ### Space Savings Scale

package/docs/SEARCH-ENGINE-V2.2.0.md

@@ -418,6 +418,7 @@ weights = {'bm25': 0.4, 'semantic': 0.3, 'graph': 0.3}  # Default
 - Index time is one-time cost
 - Search time scales sub-linearly (inverted index efficiency)
 - Hybrid search includes fusion overhead (~10-15ms)
+- These are projected estimates for the optional BM25 engine. See wiki Performance Benchmarks for measured end-to-end search latency.
 
 ---
 

package/hooks/post-recall-hook.js

@@ -0,0 +1,53 @@
+#!/usr/bin/env node
+/**
+ * SuperLocalMemory V2 - Post-Recall Hook (v2.7.4)
+ * Copyright (c) 2026 Varun Pratap Bhardwaj
+ * Licensed under MIT License
+ *
+ * Claude Code hook that tracks recall events for implicit signal collection.
+ * This hook fires after the slm-recall skill completes, recording timing
+ * data that the signal inference engine uses to detect satisfaction/dissatisfaction.
+ *
+ * Installation: Automatically registered via install-skills.sh
+ * All data stays 100% local.
+ */
+
+const fs = require('fs');
+const path = require('path');
+const os = require('os');
+
+const MEMORY_DIR = path.join(os.homedir(), '.claude-memory');
+const HOOK_LOG = path.join(MEMORY_DIR, 'recall-events.jsonl');
+
+// Parse input from Claude Code hook system
+function main() {
+    try {
+        const timestamp = Date.now();
+        const args = process.argv.slice(2);
+
+        // Extract query from args if available
+        let query = '';
+        for (let i = 0; i < args.length; i++) {
+            if (args[i] && !args[i].startsWith('--')) {
+                query = args[i];
+                break;
+            }
+        }
+
+        if (!query) return;
+
+        // Append recall event to JSONL log (lightweight, append-only)
+        const event = JSON.stringify({
+            type: 'recall',
+            query: query.substring(0, 100), // Truncate for privacy
+            timestamp: timestamp,
+            source: 'claude-code-hook',
+        });
+
+        fs.appendFileSync(HOOK_LOG, event + '\n', { flag: 'a' });
+    } catch (e) {
+        // Hook failures must be silent
+    }
+}
+
+main();