Unified Memory Skill

Name: unified-memory
Rating: 76
Author: AmitabhainArunachala

Three-layer memory architecture:

•Canonical Layer — Structured SQLite storage with FTS
•Mem0 Layer — Vector embeddings via sqlite-vec for semantic search
•Strange Loop Layer — Self-referential memory graph with emergent insights

Quick Start

bash

# Initialize database
python3 scripts/init_memory.py

# Capture a memory
./memory-cli.sh capture \
  --type insight \
  --importance 9 \
  --content "Unified memory enables perfect recall across all knowledge" \
  --tags "memory,architecture,breakthrough"

# Search (hybrid: text + semantic)
./memory-cli.sh search "knowledge architecture" --limit 10

# Find emergent insights
./memory-cli.sh insights

# Get related memories
./memory-cli.sh related <memory_id>

Architecture

code

┌─────────────────────────────────────────────────────────────┐
│                     UNIFIED MEMORY                           │
├─────────────────────────────────────────────────────────────┤
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐       │
│  │   STRANGE    │  │   CANONICAL  │  │     MEM0     │       │
│  │    LOOP      │◄──┤   MEMORY     │◄──┤    LAYER    │       │
│  │   LAYER      │  │   (SQLite)   │  │  (Vector)    │       │
│  └──────────────┘  └──────────────┘  └──────────────┘       │
│         │                   │                   │            │
│         └───────────────────┼───────────────────┘            │
│                             │                                │
│                    ┌────────┴────────┐                      │
│                    │  MEMORY MANAGER │                      │
│                    │   (Unified API) │                      │
│                    └─────────────────┘                      │
└─────────────────────────────────────────────────────────────┘

Commands

Command	Description
`capture`	Store new memory across all layers
`search`	Hybrid text/semantic search
`recall`	Rich memory with context
`related`	Find connected memories
`insights`	Generate emergent insights
`context`	Get memories for current task
`consolidate`	Weekly memory consolidation
`reflect`	System self-analysis
`bridge`	Sync with memory-system-v2

Python API

python

from unified_memory import MemoryManager

mm = MemoryManager()

# Capture
memory_id = mm.capture(content="Insight", importance=9, tags=["ai"])

# Search
results = mm.search("consciousness research", search_type="hybrid")

# Context for task
context = mm.get_context_for_task("Writing R_V paper", max_memories=5)

Integration

•memory-system-v2: Bidirectional sync via bridge command
•Obsidian PKM: Vault queries via unified interface
•RLM: Deep synthesis when semantic search insufficient

Performance

•Search: <20ms (sqlite-vec)
•Capture: <50ms
•Consolidation: Background async

See references/unified_memory_proposal.md for full design.