hivellm/Memory

<!-- MEMORY:START -->

Persistent Memory System

The rulebook memory system provides persistent context across AI sessions using hybrid search (BM25 keyword + HNSW vector) with zero native dependencies.

3-Layer Search Pattern (Token-Efficient)

Layer 1 — Compact Search: Get brief results to scan relevance.

rulebook_memory_search({ query: "authentication bug", mode: "hybrid", limit: 10 })
→ Returns: { id, title, type, score, matchType, createdAt } per result (~50 tokens each)

Layer 2 — Timeline: Get chronological context around a memory.

rulebook_memory_timeline({ memoryId: "abc-123", window: 5 })
→ Returns: 5 memories before + anchor + 5 memories after (~200 tokens each)

Layer 3 — Full Details: Get complete content only for selected memories.

rulebook_memory_get({ ids: ["abc-123", "def-456"] })
→ Returns: Full memory objects with content (~500-1000 tokens each)

Memory Types

• bugfix: Bug fixes, error resolutions
• feature: New features, additions
• refactor: Code restructuring
• decision: Architectural decisions (protected from eviction)
• discovery: Insights and learnings
• change: Updates and modifications
• observation: General observations

CLI Commands

rulebook memory search "authentication bug"     # Hybrid search
rulebook memory save "Decided to use sql.js" --type decision --title "DB Choice"
rulebook memory list --limit 10                  # Recent memories
rulebook memory stats                            # Database statistics
rulebook memory cleanup --force                  # Force eviction
rulebook memory export --format json             # Export all memories

Configuration (.rulebook)

{
  "memory": {
    "enabled": true,
    "dbPath": ".rulebook-memory/memory.db",
    "maxSizeBytes": 524288000,
    "vectorDimensions": 256
  }
}

Privacy

Content between <private> and </private> tags is automatically stripped before storage.

<!-- MEMORY:END -->