agentic-flow 2.0.1-alpha.16 → 2.0.1-alpha.18

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -2,6 +2,86 @@
2
2
 
3
3
  All notable changes to this project will be documented in this file.
4
4
 
5
+ ## [2.0.1-alpha.18] - 2025-12-31
6
+
7
+ ### Added
8
+ - **Persistent SQLite Embedding Cache**: Cross-session embedding persistence
9
+ - New `EmbeddingCache` class with SQLite (better-sqlite3) backend
10
+ - **9000x faster** cached lookups: ~0.045ms vs ~400ms ONNX inference
11
+ - Batch writes with transactions: 0.016ms per embedding
12
+ - Stored at `~/.agentic-flow/embedding-cache.db`
13
+ - LRU eviction with configurable max entries (default: 10,000)
14
+ - Automatic cleanup of entries older than 30 days
15
+ - WAL mode for concurrent access
16
+
17
+ - **Cache Hierarchy** (checked in order):
18
+ 1. In-memory LRU cache (~0.001ms) - fastest, ephemeral
19
+ 2. Persistent SQLite cache (~0.045ms) - persists across sessions
20
+ 3. ONNX inference (~400ms) - only on complete cache miss
21
+
22
+ - **EmbeddingService Cache Methods**:
23
+ - `clearPersistentCache()` - Clear SQLite cache
24
+ - `clearAllCaches()` - Clear both memory and persistent caches
25
+ - `getPersistentCacheStats()` - Get cache statistics
26
+
27
+ - **Environment Variables**:
28
+ - `AGENTIC_FLOW_PERSISTENT_CACHE=true|false` (default: true)
29
+
30
+ ### Performance (with persistent cache)
31
+ | Operation | Time | Notes |
32
+ |-----------|------|-------|
33
+ | ONNX cold | ~400ms | First inference + cache write |
34
+ | SQLite read | ~0.045ms | Persistent cache lookup |
35
+ | SQLite batch write | 0.016ms/ea | Transactional batch insert |
36
+ | Memory cache | ~0.001ms | In-memory LRU lookup |
37
+ | **Speedup** | **9000x** | SQLite vs ONNX inference |
38
+
39
+ ### Changed
40
+ - EmbeddingService now checks persistent cache before ONNX inference
41
+ - Only semantic (ONNX) embeddings are persisted (simple hash-based are not)
42
+
43
+ ## [2.0.1-alpha.17] - 2025-12-31
44
+
45
+ ### Added
46
+ - **10 Attention Mechanisms** (ruvector@0.1.63):
47
+ - `MultiHeadAttention` - Pattern matching
48
+ - `FlashAttention` - Memory-efficient O(n)
49
+ - `HyperbolicAttention` - Hierarchical data (AST, trees)
50
+ - `LinearAttention` - O(n) real-time hooks
51
+ - `LocalGlobalAttention` - Long document summarization
52
+ - `MoEAttention` - Multi-agent routing
53
+ - `GraphRoPeAttention` - Code structure analysis
54
+ - `EdgeFeaturedAttention` - Import/dependency graphs
55
+ - `DualSpaceAttention` - Euclidean + Hyperbolic
56
+
57
+ - **Extended Worker Pool Operations**:
58
+ - `speculativeEmbed(files)` - Pre-embed for post-edit
59
+ - `analyzeAST(files)` - Multi-file parallel AST
60
+ - `analyzeComplexity(files)` - Cyclomatic metrics
61
+ - `buildDependencyGraph(files)` - Import graph
62
+ - `securityScan(files)` - Parallel SAST
63
+ - `ragRetrieve(query, chunks)` - Parallel RAG
64
+ - `rankContext(query, contexts)` - Relevance ranking
65
+ - `deduplicate(texts)` - Semantic deduplication
66
+ - `gitBlame(files)` - Parallel blame analysis
67
+ - `gitChurn(patterns, since)` - Code churn metrics
68
+
69
+ - **Parallel Attention Compute**:
70
+ - `parallelAttentionCompute()` - 3-4x faster multi-query
71
+ - `batchAttentionCompute()` - Batch Q-K-V sets
72
+ - `computeFlashAttentionAsync()` - Non-blocking flash
73
+ - `getAttentionForUseCase()` - Auto-select attention type
74
+
75
+ - **Training Utilities**:
76
+ - `AdamOptimizer` - Attention parameter training
77
+ - `infoNceLoss()` - Contrastive learning
78
+ - `mineHardNegatives()` - Hard negative mining
79
+ - `benchmarkAttention()` - Performance comparison
80
+
81
+ ### Changed
82
+ - Updated ruvector dependency: 0.1.62 → 0.1.63
83
+ - Intelligence bridge supports all attention mechanisms
84
+
5
85
  ## [2.0.1-alpha.16] - 2025-12-31
6
86
 
7
87
  ### Added